Version 1.0 · Released September 14, 2026

Meta-Analysis Studio

Prepare effect sizes, fit fixed- or random-effects models, examine heterogeneity, and generate publication-ready outputs in one browser-based workflow.

Download empty CSV template

Calculations run entirely in your browser No account required Files are not uploaded to ERN Institute

Studio workflow

Four steps from study results to finished outputs

Loading a file completes the preparation automatically. You run the pooled meta-analysis only after reviewing the prepared data.

Step 1 Load study results Upload an ERN data file or use the built-in example.
Step 2 Review automatically prepared data Effect sizes, validation checks, and unpooled forest plots are created as soon as the file loads.
Step 3 Configure and run the meta-analysis Choose the effect-size metric, model, and settings, then run the pooled analysis.
Step 4 Review result and export Interpret the pooled estimate and forest plot, then download publication-ready files.

Step 1

Load study results

Load the ERN template and the Studio will immediately calculate the effect sizes, validate every row, and generate separate unpooled forest plots for each effect-size metric. No separate run command is required for preparation. The pooled meta-analysis does not run until Step 3, and no data are sent to ERN.

Download example CSV

Created automatically when the file loads Calculated effect sizes · row-by-row validation · unpooled forest plots grouped by metric

Choose a CSV or drop it here Maximum file size: 5 MB

Load your CSV or try the built-in example. The Studio will create the Step 2 effect sizes, validation report, and unpooled forest plots automatically.

Step 2 Review automatically prepared data

Open review

Automatically calculated effect sizes

Loading the file created these effect sizes, validation checks, and unpooled forest plots without an additional command. This panel opens automatically only when errors block pooling. Warnings and calculation details remain available here for review.

The workbook preserves every loaded row, records calculation assumptions, explains validation issues, and includes fitted models and figures after an analysis is run.

Calculated effect sizes

Open this table when you want to inspect the calculated effects, variances, confidence intervals, or row-level warnings.

Study	Effect ID	Metric	Effect	Variance	SE	95% CI low	95% CI high	Natural metric	Natural effect	Warning

Step 3

Configure and run the meta-analysis

Choose an effect-size metric and model. Default statistical settings are applied automatically and can be changed under Advanced settings.

Effect-size metric Analysis model

Current settings Default settings

REML · automatic inference · 95% confidence interval

A prediction interval is shown automatically when at least five effects are available.

Advanced settings

Between-study variance Inference Confidence level Prediction interval

Include a common-effect sensitivity analysis. Optional comparison shown in the on-page results and publication workbook.

Each effect-size metric produces its own pooled estimate; different metrics are never combined.

Upload data to configure an analysis.

Step 4

Review the result

Pooled forest plot

Study weights reflect the selected model. The diamond shows the pooled confidence interval; a prediction interval is shown when requested and supported.

Statistical details

Study weights and model data

Weights sum to 100% within the selected model.

Study	Effect ID	Effect	CI low	CI high	Weight

Suggested reporting text

Suggested methods text

Suggested results text

More downloads

Method definitions and safeguards

Common-effect (fixed-effect): inverse-variance weights assume that all included estimates target one common underlying effect.

Random-effects: weights incorporate estimated between-study variance. REML is the default; Paule–Mandel is retained as a strong sensitivity alternative, and DerSimonian–Laird is labeled as a legacy method for reproducing older analyses.

Heterogeneity: for random-effects models, I² and H² are calculated from the selected τ² estimator and a typical within-study variance so the displayed heterogeneity measures remain internally consistent. Cochran's Q is reported separately.

Automatic inference: the Studio uses Knapp–Hartung when a random-effects model has more than two effects and the estimated between-study variance (τ²) is positive after applying a numerical zero tolerance; otherwise it uses Wald inference. This is not a statistical significance test of heterogeneity. The exact decision is generated from the fitted model and recorded in the export.

Prediction interval: automatic mode displays an interval when at least five effects are available. The interval uses the same critical-value framework as the fitted model: t(k−1) under Knapp–Hartung and normal critical values under Wald. Advanced settings can request it with three or four effects, accompanied by a caution.

Dependency safeguard: repeated study labels trigger a confirmation step. Confirmation does not solve statistical dependence; use it only when the rows are genuinely independent.

Supported inputs

Common inputs

Independent meansMeans, SDs, and sample sizes → Hedges g

Independent tStudent t and group sizes → Hedges g

Independent FTwo-group F(1, df₂), group sizes, and direction → Hedges g

CorrelationPearson r and n → Fisher z

Binary outcomeEvents and totals → log odds ratio

Generic inverse-varianceEffect estimate plus sampling variance or standard error → analysis-ready effect

Calculation definitions and current limits

Independent means: pooled-standard-deviation Cohen d, corrected to Hedges g. The reported variance includes the small-sample correction.

Independent t: converts a Student independent-groups t statistic to Cohen d, then applies the Hedges correction. Welch, paired, repeated-measures, adjusted-model, and mixed-model t statistics are not supported by this pathway.

Independent F: supports only a simple two-group between-subjects test with numerator df = 1 and denominator df = n₁ + n₂ − 2. Because F has no sign, f_direction is required. Omnibus, ANCOVA, repeated-measures, multilevel, and mixed-model F statistics must not be entered here.

Correlations: reports Fisher z for analysis and Pearson r as the natural-scale value.

Binary outcomes: reports the log odds ratio for analysis and the odds ratio on its natural scale. A 0.5 continuity correction is disclosed when an otherwise informative 2×2 table contains a zero cell. Studies with no events in either group, or all events in both groups, are retained in the audit trail but excluded from odds-ratio pooling because they contain no comparative information.

Generic inverse-variance: accepts an already calculated effect size on a named analysis scale plus either its sampling variance or standard error. This is the appropriate route when a study design has already been handled with a valid external method and should not be forced into one of the raw-data conversion pathways.

Current limit: raw-data conversion does not support paired, repeated-measures, cluster-randomized, multilevel, adjusted-model, omnibus-F, survival, or non-independent effect structures. If a valid effect estimate and variance have already been calculated for one of these designs, use generic inverse-variance input. Non-independent effects still require a dependency-aware model and should not be pooled as independent.