Contents
On this page
- Scope and limits of software choice in meta-analysis
- Package defaults that influence pooled estimates
- Intended applications of the principal software packages
- Bayesian and extended meta-analytic models
- Recommended changes to default settings and associated thresholds
- Criteria for selecting meta-analysis software
- Reporting the software and settings used
- Limitations that software cannot address
- Key points
- Application to your study
- Frequently asked questions
Academy · Research Data Analytics & AI · Meta-analysis software
Software for Meta-Analysis: The Influence of Package Defaults on Pooled Estimates
The choice of software rarely determines the quality of a synthesis. Software influences results principally through its defaults — the between-study variance estimator, the confidence interval method and the weighting of small trials — and these defaults differ between packages in ways that matter most when few studies are available. The tool should therefore be selected according to the analysis required, and the version, the function and every changed setting should be reported.
The choice of meta-analysis software is frequently considered only after data extraction is complete, with the expectation that an appropriate package will allow a limited evidence base to yield a publishable result; no package can achieve this. The influence of a package is narrower but more consequential than is commonly assumed: its defaults implicitly select an estimator and an interval method, and these choices diverge most where reviews are weakest. This article examines where software packages differ, the intended application of each, and how to report software and settings reproducibly.
Learning objectives
- The aspects of a result that are and are not affected by the choice of software
- The three defaults that differ between packages and may alter a conclusion: tau-squared estimator, interval method, and continuity corrections
- The intended application of each of the main tools — RevMan, metafor, Stata’s
metanandmetaan, CMA, JASP, Meta-Essentials, MetaWin, OpenMEE - Which packages support Bayesian, multivariate and structural-equation extensions, and the circumstances in which these are required
- The thresholds Veroniki and colleagues recommend — HKSJ from five studies, sensitivity analysis across methods below ten
- How to write a reproducible statement of the software and settings used
Common misconceptions
- “A better package will give a better analysis.” The quality of a synthesis is determined by the review question, the search and the data extraction. No package compensates for deficiencies in any of these.
- “All meta-analysis software gives the same answer.” On the same data, defaults differ in the estimator of between-study variance, the confidence interval method and the continuity corrections for zero cells. The divergence is largest when studies are few.
- “Naming the software is enough for reproducibility.” The statement “Analyses were conducted in R” does not identify the analysis. A reader requires the package, version, function and non-default arguments.
- “Point-and-click tools are not credible.” RevMan, CMA, JASP and Meta-Essentials are all documented in the peer-reviewed literature. Credibility depends on whether the analyses performed were reported, not on whether they were scripted.
- “The default is the recommended option.” Defaults reflect historical practice. The DerSimonian and Laird estimator is the most widely implemented default and is known to produce intervals that are too narrow in small reviews.
Scope and limits of software choice in meta-analysis
The fundamental limitation should be stated first. The factors that determine whether a meta-analysis is worth reading (whether the studies address one answerable question, whether the search was adequate, whether the extraction was accurate and duplicated, and whether risk of bias was appraised and used) are all settled before any software is opened. A review built on an inconsistent extraction will produce an erroneous estimate with apparent precision in every available package.
The aspects that software genuinely determines fall into three groups. The first is defaults: which between-study variance estimator, which confidence interval method and how zero cells are handled. These choices change the reported numbers, and most users never see them being made. The second is capability: whether a network meta-analysis, a multivariate model, a Bayesian model with an informative prior on heterogeneity, or a dose-response analysis can be fitted at all. The third is auditability: whether the analysis exists as a script that another analyst can re-run, or as a sequence of menu selections that must be recalled.
The third consideration deserves more weight than it usually receives. A scripted analysis can be corrected, extended during revision and reproduced by a co-author. An analysis conducted through menu selections usually cannot, which is why the practical argument for a code-based tool concerns the demands of peer review more than statistics. The systematic review page covers the upstream steps, which are more important than any of these considerations.
Package defaults that influence pooled estimates
The between-study variance estimator. DerSimonian and Laird’s moment estimator, introduced in Controlled Clinical Trials, is the historical default in most packages 42. Alternatives include restricted maximum likelihood, the Sidik and Jonkman estimator published in Applied Statistics, the Biggerstaff and Tweedie approach, and the Bayesian treatment in Higgins, Thompson and Spiegelhalter’s re-evaluation in the Journal of the Royal Statistical Society. Viechtbauer compared their bias and efficiency in the Journal of Educational and Behavioral Statistics 24. Different estimators yield different values of tau-squared, and therefore different weights and a different pooled estimate.
The confidence interval method. Veroniki and colleagues, in Research Synthesis Methods, identified 15 confidence interval methods for the random-effects summary effect and grouped them into 7 categories 44. The Wald interval that most defaults produce is too narrow when studies are few; the Hartung-Knapp-Sidik-Jonkman approach, from Hartung and Knapp in Statistics in Medicine and Sidik and Jonkman, with the few-studies behaviour examined by Röver, Knapp and Friede in BMC Medical Research Methodology, is the usual recommendation 13,31.
Zero-cell handling. For binary outcomes, packages apply continuity corrections differently, and some exclude double-zero studies by default. With rare events, this choice can change the direction of the result, and it is almost never reported.
None of these issues is unusual. Two competent analysts, given identical extracted data and asked for a random-effects meta-analysis, will report different intervals if they use different tools without specifying these three settings, and neither will have made an error 22,41.
Intended applications of the principal software packages
The table below groups the main options by intended function rather than by price or interface. Brief notes follow on the tools most often misjudged.
RevMan exists to produce Cochrane reviews, and the statistical methods it implements are those described by Deeks, Higgins and Altman in the Cochrane Handbook 1,2. It is well suited to a standard pairwise review of randomised trials and is deliberately limited beyond that; selecting it and subsequently requiring a method it does not offer is a common and avoidable problem.
metafor, documented by Viechtbauer in the Journal of Statistical Software, is the broadest general-purpose option, offering most estimators and interval methods, meta-regression, multilevel and multivariate models, and the influence diagnostics that Viechtbauer and Cheung set out in Research Synthesis Methods 17,18. Where the requirements of an analysis are not yet known, it is the most secure choice.
Stata has a long documented history in this field. Sharp and Sterne’s original command and Bradburn, Deeks and Altman’s alternative appeared in the Stata Technical Bulletin; Harris and colleagues documented metan for fixed- and random-effects meta-analysis across pages 3 to 28 of The Stata Journal volume 8, number 1, and Kontopantelis and Reeves documented metaan in volume 10, number 3, pages 395 to 407 19,21,39,44. For groups already working in Stata, there is no methodological reason to change.
Comprehensive Meta-Analysis is the commercial tool most closely tied to a textbook treatment, Borenstein, Hedges, Higgins and Rothstein’s Introduction to Meta-Analysis, which makes it straightforward to learn and to report, provided that the selected options are recorded 20,25.
Meta-Essentials, introduced and validated by Suurmond, van Rhee and Hak in Research Synthesis Methods, is a free spreadsheet-based set of workbooks 10. It is a reasonable choice for a straightforward pairwise analysis and for teaching, and its validation paper is the appropriate citation.
The main tools, and what each is designed to do
| Tool | What it is | Best suited to | Documented in |
|---|---|---|---|
| RevMan | Cochrane’s review management and analysis environment | Standard pairwise reviews of randomised trials, in the Cochrane workflow | Deeks, Higgins and Altman, Cochrane Handbook for Systematic Reviews of Interventions |
| metafor (R) | General-purpose meta-analysis package: most estimators and interval methods, meta-regression, multilevel and multivariate models, diagnostics | Any analysis whose requirements may grow during revision | Viechtbauer, Journal of Statistical Software; diagnostics with Cheung in Research Synthesis Methods |
Stata: metan, metaan | Fixed- and random-effects commands with a long documented history | Groups already working in Stata; no methodological reason to migrate | Harris et al, The Stata Journal 8(1):3-28; Kontopantelis and Reeves, The Stata Journal 10(3):395-407 |
| Comprehensive Meta-Analysis | Commercial point-and-click package tied closely to a standard textbook treatment | Users who want a short learning curve and will record their option selections | Borenstein, Hedges, Higgins and Rothstein, Introduction to Meta-Analysis |
| Meta-Essentials | Free spreadsheet workbooks for pairwise meta-analysis | Straightforward analyses and teaching | Suurmond, van Rhee and Hak, Research Synthesis Methods |
| JASP | Free graphical program with frequentist and Bayesian meta-analysis modules | A Bayesian sensitivity analysis without writing code; publication-bias adjustment | Love et al, Journal of Statistical Software; Berkhout et al, Behavior Research Methods; Bartoš et al, AMPPS |
| bayesmeta, brms, WinBUGS (R / standalone) | Bayesian random-effects, general multilevel, and general-purpose MCMC | Few studies with defensible prior information; models outside the standard set | Röver, Journal of Statistical Software; Bürkner, Journal of Statistical Software; Lunn et al, Statistics and Computing |
| metaSEM, BFpack (R) | Meta-analysis via structural equation modelling; Bayes factor testing of constrained hypotheses | Multivariate and three-level models; hypothesis comparison rather than null testing | Cheung, Frontiers in Psychology; Mulder et al, Journal of Statistical Software |
| MetaWin 3, OpenMEE | Open-source tools developed for ecology and evolutionary biology | Non-biomedical effect measures and designs | Rosenberg, Frontiers in Bioinformatics; Wallace et al, Methods in Ecology and Evolution |
Each tool is characterised from its own peer-reviewed documentation, cited in the final column. Page and volume details for the Stata commands are as published. No comparative evaluation of these tools against one another is cited here, because the published literature does not support ranking them for review quality.
Bayesian and extended meta-analytic models
Bayesian meta-analysis justifies the additional effort in one specific situation: when studies are few, so that the between-study variance is poorly estimated and an informative prior on it genuinely contributes information rather than merely imposing an opinion. Röver’s bayesmeta package, documented in the Journal of Statistical Software, is designed for this situation and has a companion treatment of Bayesian random-effects meta-regression with Friede in Computer Methods and Programs in Biomedicine 5,8. brms, documented by Bürkner in the Journal of Statistical Software, is the general multilevel option and extends well beyond standard pairwise models 11. WinBUGS, described by Lunn and colleagues in Statistics and Computing, is the older general-purpose sampler still visible in published network meta-analyses 36.
JASP merits mention because it makes Bayesian analysis available without code. Love and colleagues documented the program in the Journal of Statistical Software; Berkhout and colleagues published a tutorial on Bayesian model-averaged meta-analysis in Behavior Research Methods; and Bartoš and colleagues, in Advances in Methods and Practices in Psychological Science, cover adjusting for publication bias in JASP and R with selection models, PET-PEESE and robust Bayesian meta-analysis 4,6,9. For a researcher seeking a defensible Bayesian sensitivity analysis alongside a frequentist main analysis, this is the most direct option. The Bayesian random-effects page covers the model itself.
Beyond pairwise: metaSEM, documented by Cheung in Frontiers in Psychology, handles meta-analysis through structural equation modelling, which is the route to multivariate and three-level models 14. BFpack, from Mulder and colleagues in the Journal of Statistical Software, handles Bayes factor testing of constrained hypotheses 7. Network meta-analysis is addressed specifically in the network meta-analysis hub.
Outside biomedicine, MetaWin 3, released by Rosenberg in Frontiers in Bioinformatics, and OpenMEE, from Wallace and colleagues in Methods in Ecology and Evolution, are the established open-source options in ecology and evolutionary biology, where effect measures and study designs differ sufficiently that the medical packages are poorly suited 3,12.
Recommended changes to default settings and associated thresholds
Veroniki and colleagues, in Research Synthesis Methods, give two usable rules from their comparison of the 15 interval methods 44. They recommend the Hartung-Knapp-Sidik-Jonkman method as standard for meta-analyses of five or more studies, and sensitivity analysis across two or three methods when fewer than ten studies are pooled 44. Both are changes of settings rather than of package: metafor, Stata and the Cochrane tools can all implement them, and most will not do so unless instructed.
Two further settings warrant a deliberate decision rather than acceptance of the default. The first is the estimator of between-study variance, for which restricted maximum likelihood is a common modern default and the moment estimator the historical one. The second is whether an interval around tau-squared itself is reported. Viechtbauer, in Statistics in Medicine, developed a method that maintains nominal coverage probabilities even in small samples when model assumptions hold, and had the most accurate coverage in simulations where normality and known sampling variances hold only asymptotically; it is implemented but rarely used.
Finally, the nature and limits of the heterogeneity output should be kept in view. Higgins and Thompson, in Statistics in Medicine, proposed three statistics (H, I-squared and R) to describe heterogeneity independently of the effect metric and the number of studies, illustrated on five datasets 30. Packages print I-squared by default. They do not indicate that I-squared is a proportion rather than an amount, and they do not prevent its use as a justification for pooling. The tests themselves are covered in the statistical tests in meta-analysis page.
Source: Veroniki AA, Jackson D, Bender R, Kuss O, Langan D, Higgins JPT, Knapp G, Salanti G. Methods to calculate uncertainty in the estimated overall effect size from a random-effects meta-analysis. Research Synthesis Methods. The review recommends sensitivity analysis across two or three methods when fewer than ten studies are pooled.
Criteria for selecting meta-analysis software
A standard pairwise review of randomised trials for a Cochrane-style output. RevMan, following the Cochrane Handbook methods, is sufficient.
An analysis that may require meta-regression, multilevel structure, unusual effect measures or diagnostics during revision. metafor is appropriate, or Stata for groups that already work in it. Discovering during revision that a tool cannot fit a model requested by a reviewer is costly.
Few studies, or a genuine reason to use prior information. A Bayesian sensitivity analysis should be added using bayesmeta, brms or JASP. It should be reported as a sensitivity analysis unless the Bayesian model is the primary analysis, in which case this should be stated and the priors justified.
An analysis that others must be able to re-run. In this case, the deciding factor is whether the analysis is scripted. This is usually the consideration that should decide between otherwise equivalent options, and it is the one least often considered.
One negative finding should be stated: there is no evidence that choosing one mainstream package over another improves the quality or publishability of a review. Claims to that effect derive from vendors and from habit, not from the methodological literature.
Reporting the software and settings used
Most published software statements are inadequate for reproduction. The test is simple: whether a competent analyst with the same extracted data could reproduce the reported numbers from the methods section alone. If the estimator, the interval method and the version are missing, this is not possible.
The same analysis, described two ways
Not reproducible
All statistical analyses were performed using R software. A random-effects model was used to pool the data. Heterogeneity was assessed using the I-squared statistic. Publication bias was assessed using funnel plots and Egger’s test. A p-value below 0.05 was considered statistically significant.
Which package, which version? “R software” is an environment, not an analysis. Defaults differ between packages and between versions of the same package.
Which estimator, which interval? These are unstated, leaving 15 interval methods in 7 categories open, and the methods diverge most in small reviews.
Zero cells handled how? This is not mentioned, although for rare events it can change the direction of the pooled result.
Egger’s test on how many studies? If the review is small, the test should not have been performed, so a null result provides no reassurance.
Could anyone re-run it? No. There is no script, no version and no settings to re-run.
Reproducible
Analyses were conducted in R 4.3.1 using metafor 4.4-0, with the script and session information archived in the project repository. Log odds ratios were pooled with rma.uni using a random-effects model, with tau-squared estimated by restricted maximum likelihood (method="REML") and Hartung-Knapp-Sidik-Jonkman confidence intervals (test="knha"), chosen in advance because fewer than ten studies were expected; results using the DerSimonian and Laird estimator with Wald intervals are reported in the appendix as a sensitivity analysis, following the recommendation to compare two or three methods below ten studies. Heterogeneity is reported as tau-squared with a Q-profile confidence interval alongside I-squared, with a prediction interval for the summary effect. Double-zero studies were retained and no continuity correction was applied. Influence was assessed with influence.rma.uni and a Baujat plot. Because the review includes fewer than ten studies, no test for funnel plot asymmetry was performed.
Version numbers for both the environment and the package, with the script archived, so that the numbers can be regenerated after defaults change.
The function and non-default arguments are named, which identifies a single analysis.
The interval method is justified by the expected number of studies, and the alternative is reported as a sensitivity analysis.
An explicit decision on zero cells, so that a reader does not need to infer what the package did.
The asymmetry test is declined with a reason, which is a stronger statement than a non-significant test.
Diagnostics are named as functions, so the robustness check is inspectable rather than asserted.
The second version is approximately three times as long and contains no additional statistics; it reports the same analysis with its settings visible. Several additions follow from a threshold rather than a preference: HKSJ was chosen because fewer than ten studies were expected, and the sensitivity analysis across methods follows the same recommendation. This is the pattern to follow: the number of studies should determine the settings, and the settings should then be documented.
Limitations that software cannot address
An inconsistent extraction. Independent extraction by two reviewers detects errors; a package cannot detect that a standard deviation was entered as a standard error. This is the most common single source of an erroneous meta-analysis, and it is undetectable at later stages.
Studies that should not have been combined. No package will refuse to pool incomparable trials, and a low I-squared across clinically disparate studies provides a precise answer to an irrelevant question.
Too few studies. Software will readily produce a random-effects pooled estimate from three trials. The interval will be unreliable whichever method is selected, and the appropriate report is that the evidence base does not support a stable summary 13,44.
A biased evidence base. Trim and fill, PET-PEESE and selection models are sensitivity analyses that quantify a scenario; they do not recover studies that were never published 16,28. A result that persists only after adjustment should be described cautiously rather than defended.
Missing version information after the fact. If the package version was not recorded at the time, the numbers often cannot be reconstructed later, because defaults change between versions. The script and session information should be archived with the data, and the package should be cited in the reference list as its documentation paper requests; in the case of metafor, this is Viechtbauer’s Journal of Statistical Software article.
Key points
Key points
Software does not improve a synthesis whose question, search or extraction is weak, and no published evidence supports ranking the mainstream packages by review quality. The choice of software does determine the defaults, and these are consequential: the between-study variance estimator, the confidence interval method chosen from the 15 that Veroniki and colleagues grouped into 7 categories, and the handling of zero cells. Software should be selected for its capabilities and for whether the analysis can be preserved as a re-runnable script. The package, version, function and every non-default argument should then be reported, because without them the methods section does not describe a specific analysis.
Application to your study
Application to your study
- The analysis should be specified before the tool is selected. The model, the estimator, the interval method and the required diagnostics should be documented first, and a package that supports all of them selected subsequently.
- A scripted analysis is preferable where others will review or modify it. Revisions, co-author checks and reviewer requests are more efficiently addressed with a script than with a sequence of recalled menu selections.
- The confidence interval method should be changed deliberately. Veroniki and colleagues recommend HKSJ as standard from five studies, and sensitivity analysis across two or three methods below ten.
- The tau-squared estimator should be set explicitly, even when the default is retained. Recording the estimator documents it as a deliberate choice rather than an incidental one.
- The handling of zero cells should be specified and reported. For rare-event outcomes, this decision can change the direction of the result.
- Tau-squared should be reported with an interval, in addition to I-squared. Most packages report the proportion (I-squared) by default and the amount of heterogeneity (tau-squared) only on request.
- The software version should be recorded and the script and session information archived at the time of analysis. Defaults change between versions, and reconstructing them retrospectively is often impossible.
- The package should be cited as its documentation paper specifies. For metafor that is Viechtbauer’s Journal of Statistical Software article; most packages name a citation.
Frequently asked questions
No single package is best, and no published comparative evidence supports ranking the mainstream tools for review quality. The choice should be based on capability and reproducibility. For a standard pairwise review of randomised trials in a Cochrane workflow, RevMan implements the Cochrane Handbook methods and is sufficient. For any analysis whose requirements might expand (meta-regression, multilevel structure, unusual effect measures or influence diagnostics), metafor in R, documented by Viechtbauer in the Journal of Statistical Software, is the broadest option. For groups that work in Stata, metan and metaan are well documented and there is no methodological reason to migrate. The deciding question is usually whether the analysis will exist as a re-runnable script.
Yes, if the defaults are accepted, and the differences are largest where reviews are weakest. Three defaults vary between packages: the estimator of between-study variance, the confidence interval method for the summary effect, and the handling of zero cells in binary outcomes. Veroniki and colleagues identified 15 interval methods in 7 categories, which indicates the scale of the discretion involved. Two competent analysts given identical data and asked for “a random-effects meta-analysis” can legitimately report different intervals using different tools. For this reason, the methods section must name the estimator and the interval method, not only the software.
The environment and its version, the package and its version, the function called, and every argument changed from the default, together with the location where the script and session information are archived. The analysis-defining choices should then be stated in words: the effect scale, the between-study variance estimator, the confidence interval method and the reason for its selection, the handling of zero cells, and the diagnostics performed. “Analyses were performed in R” or “using Comprehensive Meta-Analysis” identifies nothing reproducible, because defaults differ between packages and between versions of the same package. If the version was not recorded at the time, the analysts’ own numbers often cannot be reconstructed later.
It justifies the effort in one specific case: when studies are few, so that the between-study variance is poorly estimated and an informative prior on it genuinely adds information rather than merely encoding an opinion. Röver’s bayesmeta, documented in the Journal of Statistical Software, is designed for this purpose; brms handles the general multilevel case; and JASP offers an option without code, with a tutorial on model-averaged meta-analysis by Berkhout and colleagues in Behavior Research Methods. Unless the Bayesian model is the primary analysis, it should be reported as a sensitivity analysis with the priors explicitly justified; an unjustified prior on heterogeneity is the objection that reviewers raise.
RevMan is sufficient for its intended purpose: a standard pairwise meta-analysis of randomised trials using the methods that Deeks, Higgins and Altman set out in the Cochrane Handbook. It becomes problematic when a reviewer requests an analysis outside that set (a different interval method, a multilevel model, influence diagnostics or an unusual effect measure) for which it provides no option. Where this is a realistic possibility, beginning in metafor or Stata entails a learning curve at the outset but avoids a re-analysis later. The converse error, abandoning a functioning Cochrane workflow for R without good reason, should also be avoided.
No, and the claim is not supported anywhere in the methodological literature. Whether a synthesis is robust depends on whether the included studies address one answerable question, whether the search was adequate and documented, whether extraction was duplicated and accurate, and whether risk of bias was appraised and used in the synthesis. All of these are settled before any software is opened. A package will pool an inconsistent extraction without warning and produce a precise-looking interval around an erroneous estimate. The reproducibility of the reporting is the one area in which tooling genuinely affects how a reviewer perceives the work.
Three, as a minimum. The confidence interval method should be set rather than a Wald interval accepted: Veroniki and colleagues recommend Hartung-Knapp-Sidik-Jonkman as standard for meta-analyses of five or more studies, with sensitivity analysis across two or three methods when fewer than ten studies are pooled. The between-study variance estimator should be set explicitly, even if the package default is retained, so that the choice is recorded. The handling of zero cells in binary outcomes should also be decided, since packages differ and, with rare events, the choice can change the direction of the pooled estimate. In addition, a confidence interval around tau-squared should be requested, since most tools produce it only on request.
Network meta-analysis requires a tool designed for it, and this is one of the clearest cases in which capability rather than familiarity should determine the choice. R packages designed for network models, and general-purpose Bayesian samplers such as WinBUGS, described by Lunn and colleagues in Statistics and Computing, are the established options; brms and metaSEM also extend to multivariate structures. RevMan does not fit network models. Beyond the software question, network meta-analysis has reporting and assumption requirements (transitivity, consistency and network geometry) that are substantially more demanding than those of pairwise analysis, and the network meta-analysis hub covers them.
References
- Deeks JJ, Higgins JPT, Altman DG. Analysing data and undertaking meta-analyses. Cochrane Handbook for Systematic Reviews of Interventions. 2019:241-284. doi:10.1002/9781119536604.ch10 read 9 October 2026. doi.org/10.1002/9781119536604.ch10
- The Nordic Cochrane Centre. Review Manager (RevMan) version 4.2 for Windows. The Cochrane Collaboration (software). 2003. read 9 October 2026. link.springer.com
- Rosenberg MS. Metawin 3: open-source software for meta-analysis. Frontiers in Bioinformatics. 2024;4:1305969. read 9 October 2026. arxiv.org
- Berkhout SW, Haaf JM, Gronau QF, Heck DW, Wagenmakers EJ. A tutorial on Bayesian model-averaged meta-analysis in JASP. Behavior Research Methods. 2023. read 9 October 2026. arxiv.org
- Rover C, Friede T. Using the bayesmeta R package for Bayesian random-effects meta-regression. Computer Methods and Programs in Biomedicine. 2023;229:107303. read 9 October 2026. arxiv.org
- Bartos F, Maier M, Quintana D, Wagenmakers EJ. Adjusting for publication bias in JASP and R: selection models, PET-PEESE, and robust Bayesian meta-analysis. Advances in Methods and Practices in Psychological Science. 2022;5:1-19. read 9 October 2026. arxiv.org
- Mulder J, Williams DR, Gu X, Tomarken A, Boing-Messing F, Olsson-Collentine A, Meijerink M, Menke J, van Aert R, Fox JP, Hoijtink H, Rosseel Y, Wagenmakers EJ, van Lissa C. BFpack: flexible Bayes factor testing of scientific theories in R. Journal of Statistical Software. 2021;100:1-63. read 9 October 2026. arxiv.org
- Rover C. Bayesian random-effects meta-analysis using the bayesmeta R package. Journal of Statistical Software. 2020;93(6):1-51. doi:10.18637/jss.v093.i06 read 9 October 2026. doi.org/10.18637/jss.v093.i06
- Love J, Selker R, Marsman M, Jamil T, Dropmann D, Verhagen J, Ly A, Gronau QF, Smira M, Epskamp S, Matzke D, Wild A, Knight P, Rouder JN, Morey RD, Wagenmakers EJ. JASP: Graphical statistical software for common statistical designs. Journal of Statistical Software. 2019;88:1-17. read 9 October 2026. arxiv.org
- Suurmond R, van Rhee H, Hak T. Introduction, comparison and validation of Meta-Essentials: a free and simple tool for meta-analysis. Research Synthesis Methods. 2017;8:537-553. read 9 October 2026. arxiv.org
- Burkner PC. brms: An R package for Bayesian multilevel models using Stan. Journal of Statistical Software. 2017;80:1-28. read 9 October 2026. arxiv.org
- Wallace BC, Lajeunesse MJ, Dietz G, Dahabreh IJ, Trikalinos TA, Schmid CH, Gurevitch J. OpenMEE: intuitive, open-source software for meta-analysis in ecology and evolutionary biology. Methods in Ecology and Evolution. 2016. read 9 October 2026. arxiv.org
- Rover C, Knapp G, Friede T. Hartung-Knapp-Sidik-Jonkman approach and its modification for random-effects meta-analysis with few studies. BMC Medical Research Methodology. 2015;15:99. doi:10.1186/s12874-015-0091-1 read 9 October 2026. doi.org/10.1186/s12874-015-0091-1
- Cheung MWL. metaSEM: An R package for meta-analysis using structural equation modeling. Frontiers in Psychology. 2015;5. read 9 October 2026. arxiv.org
- Riley RD, Higgins JPT, Deeks JJ. Interpretation of random effects meta-analyses. BMJ. 2011;342:d549. doi:10.1136/bmj.d549 read 9 October 2026. doi.org/10.1136/bmj.d549
- Sterne JAC, Sutton AJ, Ioannidis JPA, Terrin N, Jones DR, Lau J, Carpenter J, Rucker G, Harbord RM, Schmid CH, Tetzlaff J, Deeks JJ, Peters J, Macaskill P, Schwarzer G, Duval S, Altman DG, Moher D, Higgins JPT. Recommendations for examining and interpreting funnel plot asymmetry in meta-analyses of randomised controlled trials. BMJ. 2011;343:d4002. doi:10.1136/bmj.d4002 read 9 October 2026. doi.org/10.1136/bmj.d4002
- Viechtbauer W. Conducting meta-analyses in R with the metafor package. Journal of Statistical Software. 2010;36(3):1-48. doi:10.18637/jss.v036.i03 read 9 October 2026. doi.org/10.18637/jss.v036.i03
- Viechtbauer W, Cheung MWL. Outlier and influence diagnostics for meta-analysis. Research Synthesis Methods. 2010;1(2):112-125. doi:10.1002/jrsm.11 read 9 October 2026. doi.org/10.1002/jrsm.11
- Kontopantelis E, Reeves D. metaan: Random-effects meta-analysis. The Stata Journal. 2010;10(3):395-407. read 9 October 2026. ageconsearch.umn.edu
- Borenstein M, Hedges LV, Higgins JPT, Rothstein HR. Introduction to Meta-Analysis. 2009. read 9 October 2026. medcalc.org
- Harris RJ, Bradburn MJ, Deeks JJ, Harbord RM, Altman DG, Sterne JAC. metan: fixed- and random-effects meta-analysis. The Stata Journal. 2008;8(1):3-28. doi:10.1177/1536867x0800800102 read 9 October 2026. doi.org/10.1177/1536867x0800800102
- Bax L, Yu LM, Ikeda N, Tsuruta N, Moons KGM. A systematic comparison of software dedicated to meta-analysis of causal studies. BMC Medical Research Methodology. 2007;7:40. doi:10.1186/1471-2288-7-40 read 9 October 2026. doi.org/10.1186/1471-2288-7-40
- Bax L, Yu LM, Ikeda N, Tsuruta H, Moons KG. Development and validation of MIX: comprehensive free software for meta-analysis of causal research data. BMC Medical Research Methodology. 2006;6(1):50. doi:10.1186/1471-2288-6-50 read 9 October 2026. doi.org/10.1186/1471-2288-6-50
- Viechtbauer W. Bias and efficiency of meta-analytic variance estimators in the random-effects model. Journal of Educational and Behavioral Statistics. 2005;30(3):261-293. doi:10.3102/10769986030003261 read 9 October 2026. doi.org/10.3102/10769986030003261
- Borenstein M, Hedges L, Higgins J, Rothstein H. Comprehensive Meta-Analysis version 2. Biostat (software). 2005. read 9 October 2026. link.springer.com
- Leandro G. Meta-analysis in medical research. Blackwell Publishing / BMJ Books (book). 2005. read 9 October 2026. link.springer.com
- Bradburn MJ, Deeks JJ, Altman DG. Metan: an alternative meta-analysis command. Stata Technical Bulletin. 2003. read 9 October 2026. link.springer.com
- Steichen TJ. Nonparametric trim and fill analysis of publication bias in meta-analysis (metatrim). Stata Technical Bulletin. 2003. read 9 October 2026. link.springer.com
- Steichen TJ. Tests for publication bias in meta-analysis (metabias). Stata Journal. 2003. read 9 October 2026. link.springer.com
- Higgins JPT, Thompson SG. Quantifying heterogeneity in a meta-analysis. Statistics in Medicine. 2002;21(11):1539-1558. doi:10.1002/sim.1186 read 9 October 2026. doi.org/10.1002/sim.1186
- Hartung J, Knapp G. On tests of the overall treatment effect in meta-analysis with normally distributed responses. Statistics in Medicine. 2001;20:1771-1782. read 9 October 2026. arxiv.org
- Egger M, Davey Smith G, Altman DG. Systematic reviews in health care: meta-analysis in context. BMJ Books (book). 2001. read 9 October 2026. link.springer.com
- Sutton AJ, Lambert PC, Hellmich M, Abrams KR, Jones DR. Meta-analysis in practice: a critical review of available software. Meta-Analysis in Medicine and Health Policy (Marcel Dekker, book chapter). 2000. read 9 October 2026. link.springer.com
- Rosenberg MS, Adams DC, Gurevitch J. MetaWin: statistical software for meta-analysis version 2. Sinauer Associates (software). 2000. read 9 October 2026. link.springer.com
- Sutton AJ, Abrams KR, Jones DR, Sheldon TA, Song F. Methods for meta-analysis in medical research. Wiley (book). 2000. read 9 October 2026. link.springer.com
- Lunn DJ, Thomas A, Best N, Spiegelhalter D. WinBUGS - a Bayesian modelling framework: concepts, structure, and extensibility. Statistics and Computing. 2000;10:325-337. read 9 October 2026. ageconsearch.umn.edu
- Egger M, Sterne JAC, Davey Smith G. Meta-analysis software. BMJ. 1998;316(7126). read 9 October 2026. link.springer.com
- Bradburn MJ, Deeks JJ, Altman DG. sbe24: metan - an alternative meta-analysis command. Stata Technical Bulletin. 1998;44:4-15. read 9 October 2026. ageconsearch.umn.edu
- Sharp S, Sterne J. sbe16: Meta-analysis. Stata Technical Bulletin. 1997;38:9-14. read 9 October 2026. ageconsearch.umn.edu
- Normand SLT. Meta-analysis software: a comparative review - DSTAT version 1.10. The American Statistician. 1995;49:298-309. read 9 October 2026. link.springer.com
- Arthur W, Bennett W, Huffcutt A. Choice of software and programs in meta-analysis research: does it make a difference? Educational and Psychological Measurement. 1994;54:776-787. read 9 October 2026. link.springer.com
- DerSimonian R, Laird N. Meta-analysis in clinical trials. Controlled Clinical Trials. 1986;7(3):177-188. doi:10.1016/0197-2456(86)90046-2 read 9 October 2026. doi.org/10.1016/0197-2456(86)90046-2
- Hedges LV, Olkin I. Statistical methods for meta-analysis. 1985. read 9 October 2026. shiny.lapclin-aids.ini.fiocruz.br
- Veroniki AA, Jackson D, Bender R, Kuss O, Langan D, Higgins JPT, Knapp G, Salanti G. Methods to calculate uncertainty in the estimated overall effect size from a random-effects meta-analysis. doi:10.1002/jrsm.1319 read 9 October 2026. doi.org/10.1002/jrsm.1319
Every source above is a peer-reviewed article, a publisher’s record of one, or a standard issued by the body that owns it. Each carries the date it was last read. Where a standard moves, this article is revised rather than left up with a note.
How to cite this article
Pick your referencing style and copy it. Your university or target journal may vary the punctuation — check its guide if it has one.
Pubrica Academy. (2026, October 9). Software for Meta-Analysis: The Influence of Package Defaults on Pooled Estimates. Pubrica. https://pubrica.com/academy/concepts-definitions/software-programs-to-assist-meta-analysis/
Pubrica Academy. “Software for Meta-Analysis: The Influence of Package Defaults on Pooled Estimates.” Pubrica, 9 Oct. 2026, pubrica.com/academy/concepts-definitions/software-programs-to-assist-meta-analysis/.
Pubrica Academy (2026) Software for Meta-Analysis: The Influence of Package Defaults on Pooled Estimates. Pubrica. Available at: https://pubrica.com/academy/concepts-definitions/software-programs-to-assist-meta-analysis/ (Accessed: 15 October 2026).
Pubrica Academy. “Software for Meta-Analysis: The Influence of Package Defaults on Pooled Estimates.” Pubrica. October 9, 2026. https://pubrica.com/academy/concepts-definitions/software-programs-to-assist-meta-analysis/
Pubrica Academy, ‘Software for Meta-Analysis: The Influence of Package Defaults on Pooled Estimates’ (Pubrica, 9 October 2026) <https://pubrica.com/academy/concepts-definitions/software-programs-to-assist-meta-analysis/> accessed 15 October 2026.
Pubrica Academy. Software for Meta-Analysis: The Influence of Package Defaults on Pooled Estimates. [Internet]. Pubrica; 2026 Oct 9 [cited 2026 Oct 15]. Available from: https://pubrica.com/academy/concepts-definitions/software-programs-to-assist-meta-analysis/
Pubrica Academy, “Software for Meta-Analysis: The Influence of Package Defaults on Pooled Estimates,” Pubrica, Oct. 9, 2026. [Online]. Available: https://pubrica.com/academy/concepts-definitions/software-programs-to-assist-meta-analysis/
@misc{software_programs_to_assist_meta_analy_2026,
author = {{Pubrica Academy}},
title = {Software for Meta-Analysis: The Influence of Package Defaults on Pooled Estimates},
howpublished = {Pubrica},
year = {2026},
month = {oct},
url = {https://pubrica.com/academy/concepts-definitions/software-programs-to-assist-meta-analysis/},
note = {Accessed: <span data-cite-acc="iso">2026-10-15</span>}
}This article is written to be learned from. Cite it if you are referring to what it says; if you are using the evidence behind it, cite the sources listed above instead.
Software and settings record you can paste into your methods
This form should be completed while the analysis is run, not afterwards. Defaults change between versions, so any setting not recorded at the time may be unrecoverable later. The completed form reads as a methods paragraph with minimal editing.
META-ANALYSIS SOFTWARE AND SETTINGS RECORD
Review: ______________________ Date of analysis: ____________
Number of studies pooled: ____
1. ENVIRONMENT
Program and version: _________________________________
Package / module and version: ________________________
Operating system (if relevant to reproducibility): ___
Script location / repository: ________________________
Session information archived? [ ] yes [ ] no
2. MODEL AND FUNCTION
Function or command called: __________________________
Effect measure: ______________________________________
Analysis scale: ______________________________________
Model: [ ] fixed effect [ ] random effects
[ ] Bayesian - priors stated in section 6
3. NON-DEFAULT ARGUMENTS (list every one)
______________________________________________________
______________________________________________________
______________________________________________________
4. THE THREE SETTINGS THAT MOVE THE ANSWER
tau-squared estimator: _______________________________
[ ] this is the package default [ ] changed, because:
__________________________________________________
Confidence interval method for the summary effect: ___
[ ] HKSJ (recommended as standard from 5 studies)
[ ] Wald / DerSimonian-Laird
[ ] other: _______________________________________
Sensitivity analysis across methods?
[ ] yes - methods compared: ______________________
[ ] no - number of studies is 10 or more
Zero-cell handling (binary outcomes):
[ ] double-zero studies retained [ ] excluded
Continuity correction: ___________________________
5. HETEROGENEITY OUTPUT REQUESTED
[ ] Q with degrees of freedom
[ ] I-squared
[ ] tau-squared WITH a confidence interval (method: ______)
[ ] prediction interval for the summary effect
6. BAYESIAN ANALYSIS, IF ANY
Package: _____________________________________________
Prior on the effect: _________________________________
Prior on heterogeneity, and its justification: _______
__________________________________________________
Role: [ ] primary analysis [ ] sensitivity analysis
Convergence diagnostics reported: ____________________
7. DIAGNOSTICS RUN
[ ] leave-one-out / influence function (name it: ______)
[ ] Baujat plot
[ ] funnel plot (inspection only / with power contours)
Asymmetry test: [ ] run (which: ______) [ ] not run,
because fewer than 10 studies
8. CITATION
Package citation as its documentation requests:
__________________________________________________
[ ] added to the reference list
9. HONEST LIMITS CHECK
[ ] extraction was duplicated independently
[ ] the decision to pool was argued clinically, not from I-squared
[ ] no claim rests on an adjustment method alone
[ ] the number of studies supports the interval being reported Pubrica AcademyWritten and maintained by Pubrica’s subject matter experts. Articles are revised when the underlying standards change, and each article states the date on which it was last reviewed.
How Pubrica can help
Pubrica’s biostatisticians can identify the package that a planned analysis requires before it begins, and can set the estimator, interval method and diagnostics deliberately rather than by default. For analyses already conducted, the team can re-run the pooling across interval methods, add the prediction interval and influence diagnostics that are frequently absent from drafts, and revise the software paragraph so that another analyst could reproduce the results. Where the principal problem lies in the data extraction rather than the software, this is stated.