Meta-Analysis Module in JASP

• Meta-Analysis Module in JASP is an integrated tool that provides both classical and Bayesian meta-analytic methods through a user-friendly graphical interface.
• It supports advanced modeling techniques including multilevel, multivariate, and diagnostic accuracy analyses, effectively handling challenges like rare events and measurement error.
• The module ensures reproducibility and accessibility by automatically generating R code, high-quality visuals, and detailed workflow annotations for transparent evidence synthesis.

The Meta-Analysis module in JASP provides an integrated platform for conducting classical and Bayesian meta-analytic inference with wide-ranging methodological support and an accessible graphical user interface. This module enables researchers to synthesize results from multiple studies, estimate effect sizes and heterogeneity, perform moderator and bias analysis, and produce publication-ready outputs without requiring programming expertise. Below is an authoritative overview of the module’s capabilities, technical foundations, and its relevance to rigorous evidence synthesis.

1. Effect Size Computation and Data Preparation

JASP’s Meta-Analysis module automates the initial computation of effect sizes using a graphical interface mapping dataset columns to required input variables, such as counts or means for treatment and control groups (Bartoš et al., 11 Sep 2025). The internal engine employs routines from the metafor R package (notably metafor::escalc), providing access to effect measures relevant across empirical domains. The module supports custom effect size definitions, chained computations, and data subsetting prior to analysis.

Data preparation includes automatic conversion and labeling, guiding users through specification of the effect size metric (e.g., risk difference, odds ratio, standardized mean difference). This standardizes data for downstream meta-analytic modeling, ensuring methodological consistency and facilitating transparent analysis workflows.

2. Classical Meta-Analytic Inference and Visualization

For classical inference, the module implements state-of-the-art random-effects meta-analysis. Restricted maximum likelihood (REML) is the default estimator for the between-paper variance, and the Knapp-Hartung adjustment is applied to standard errors for improved coverage and inference accuracy (Bartoš et al., 11 Sep 2025).

Forest plots are highly configurable, displaying point estimates, confidence intervals, paper-specific information, and model-level statistics (e.g., I², H², log-likelihood, AIC, BIC). Funnel plots facilitate visual diagnosis of publication bias and small-paper effects, with centering options (null versus model estimate), power bands, and optional stratification by subgroups. The module also provides publication bias diagnostics including Egger’s regression test, Begg’s test, trim-and-fill, and fail-safe N computation.

Subgroup analyses are enabled through categorical moderator specification, with automatic stratification of estimates and omnibus tests for subgroup differences. Meta-regression extends the analysis to continuous and categorical predictors, reporting regression coefficients, omnibus slope tests (continuous predictors), and marginal mean contrasts (categorical predictors). Location-scale models and bubble plots visually represent both effect size predictions and heterogeneity variance as functions of moderators.

Cluster-robust standard error estimation is available via specification of cluster variables, employing small-sample adjustments based on the clubSandwich package for proper inference under clustering.

3. Advanced Modeling: Multilevel, Multivariate, and Diagnostic Accuracy

JASP supports multilevel meta-analytic models accounting for hierarchical data structures—multiple correlated effect sizes within studies or clusters (Bartoš et al., 11 Sep 2025). Two-level models generate separate heterogeneity estimates (e.g., τ_study, τ_study/esid), and multivariate models permit specification/import of within-paper variance-covariance matrices (with user-defined correlation patterns or direct file input).

For diagnostic accuracy meta-analysis, integration of hierarchical multinomial processing tree models allows simultaneous estimation of test sensitivity, specificity, and paper-specific prevalence under an exact likelihood, addressing limitations of normal approximations in traditional bivariate random-effects models (Guolo, 2023). The module additionally supports user selection of link functions (logit, probit, cloglog) for measure transformation and provides ROC curve summaries and confidence ellipsoids for diagnostic performance evaluation.

4. Bayesian Meta-Analysis: Estimation, Hypothesis Testing, and Model Averaging

The module offers full-featured Bayesian meta-analysis for estimating effect sizes, heterogeneity, moderator effects, and adjusting for publication bias (Bartoš et al., 11 Sep 2025).

Default priors for key parameters are μ ~ N(0,1) and τ ~ Inv-Gamma(1,0.15) for standardized mean differences. The user may customize priors or apply empirical distributions derived from large-scale evidence bases (such as the Cochrane Database) with subfield-specific hyperparameters (Bartoš et al., 2023, Bartoš et al., 2021). The Bayesian engine utilizes MCMC sampling for posterior distribution computation and Monte Carlo diagnostics (effective sample size, R̂ statistics, trace plots).

Bayesian hypothesis testing is implemented with inclusion Bayes factors via the product space method—testing presence versus absence of effect and/or heterogeneity (spike-and-slab priors). Model averaging accounts for uncertainty across fixed/random effects and model structure, with posterior estimates appropriately shrunk toward the null under weak evidence.

Bayesian meta-regression uses standardized continuous moderators and scaled orthonormal contrasts for categorical factors to ensure exchangeability and unbiased estimation of marginal effects. Multilevel modeling is supported, with the parameter ρ quantifying variance component attribution.

Publication bias is adjusted using model averaging over weight function models and PET-PEESE corrections (RoBMA‐PSMA), with posterior outputs providing diagnostic weights and evidence for pooled effects under selection bias scenarios.

5. Estimation in Challenging and Specialized Settings

The module incorporates advanced methodologies for rare event meta-analysis, measurement error, and magnitude effects:

• Sparse data and rare events are addressed using the binomial-normal hierarchical model (BNHM) with weakly informative priors (WIP) on treatment effects, obviating continuity corrections and stabilizing estimation under zero event counts (Günhan et al., 2018).
• Measurement error is handled through a flexible Bayesian hierarchical model correcting attenuation bias using paper-specific attenuation factors γ^[k], applicable to both aggregate and individual-participant data, and propagating measurement uncertainty into the posterior distributions (Campbell et al., 2020).
• For magnitude measures (absolute standardized mean differences, ASMD), squared effect modeling with appropriate weighting and heterogeneity correction yields unbiased point estimation, while interval estimation is robustly achieved by squaring endpoints of signed-effect confidence intervals (preferably using t-distribution quantiles, as in SSC_t) (Kulinskaya et al., 2023).

Additionally, the module supports estimation of mean and standard deviation from reported quantiles and other summary statistics in non-normal data, applying generalized Box-Cox transformations (Yeo–Johnson) to accommodate negative values and improve accuracy over traditional methods (Xiao et al., 2023).

6. Visualization, Reproducibility, and Accessibility

JASP’s graphical interface provides interactive effect size computation, model specification, diagnostics, and customizable visual outputs (forest plots, funnel plots, bubble plots, ROC curves). The module automatically generates corresponding R code for full reproducibility and supports multiple export formats (APA-style, LaTeX, images) for publication.

Accessibility is prioritized via progressive disclosure in the GUI—novices can conduct standard analyses, while experts access granular modeling options, prior controls, and advanced diagnostics. Workflow annotation and OSF integration ensure that meta-analytic practices are transparent and replicable.

7. Methodological Innovations and Future Directions

Recent research directions reflected in the module include:

• Monte Carlo conditioning methods for improved confidence interval coverage under small-sample random-effects meta-analysis, avoiding overconfident inference characteristic of large-sample approximations (Sugasawa et al., 2017).
• ALL-IN meta-analysis methodology, enabling anytime-valid evidence updating with type-I error control in real-time, particularly relevant for living systematic reviews and interim data synthesis (Schure et al., 2021).
• Bayesian hierarchical mean-variance modeling, incorporating Gaussian processes to flexibly capture time trends and accommodate heteroscedasticity—enabling high-precision inference in temporally heterogeneous datasets (Kubota et al., 6 Feb 2025).
• P-value function–based meta-analysis (notably Edgington’s orientation-invariant combination method), offering robust point and interval estimation with asymmetric confidence intervals under skewed data (Held et al., 15 Aug 2024).
• Integration of epistemic uncertainty through guided workflows (e.g., MetaExplorer), emphasizing structured quality assessment and sensitivity analysis in evidence aggregation, contrasting with purely statistical aggregation approaches (Kale et al., 2023).

Summary

The Meta-Analysis module in JASP constitutes a comprehensive statistical toolkit, covering classical and Bayesian inference, advanced modeling for specialized and challenging scenarios, and flexible, reproducible reporting. The integration of robust estimation techniques, dynamic visualizations, empirical priors, and methods for complex heterogeneity fundamentally supports rigorous evidence synthesis and enhances meta-analytic practice across scientific disciplines.