No results yet. Upload data and run the analysis first.

Select a moderator column in the Data Input tab to enable subgroup analysis.

Input pairwise comparison data. CSV needs columns: study, treatment_a, treatment_b, effect, variance.

▶ Run Network Meta-Analysis

Rate each study's quality. Results update automatically.

📖 Mikoshi Meta-Analysis — Complete User Guide

What is Meta-Analysis?

Meta-analysis is a statistical technique for combining the findings from independent studies. It produces a single weighted estimate of an effect size, increasing statistical power and precision beyond what any individual study can achieve.

How to Format Your CSV

Your CSV needs at minimum three columns:

• Study label — name or identifier for each study
• Effect size — the estimated effect (e.g., Cohen's d, log odds ratio, mean difference)
• Variance or Standard Error — the sampling variability for that effect
• Optional: Moderator — a categorical/continuous variable for subgroup/regression

Tab Guide

• Data — Upload CSV, map columns, run analysis
• Results — Fixed-effect and random-effects model summaries
• Forest — Forest plot, radial (Galbraith) plot, L'Abbé plot
• Funnel & Bias — Funnel plot, Egger's test, trim-and-fill, PET-PEESE, p-curve, z-curve, fail-safe N, selection model
• Diagnostics — Leave-one-out, Baujat plot, GOSH plot, influence diagnostics, outlier detection
• Cumulative — Cumulative meta-analysis sorted by year/effect/precision
• Meta-Regression — Weighted least squares regression with bubble plots
• Bayesian — Normal-normal conjugate model, prior sensitivity, Bayes factors
• Subgroups — Moderator-based subgroup analysis
• Network — Network meta-analysis for multiple treatment comparisons
• Robust — Cluster-robust variance estimation, permutation tests
• Quality — GRADE-like quality assessment, stratified analysis
• Prediction & Power — Prediction intervals, power analysis, sample size calculator
• Reporting — APA text, LaTeX, PRISMA flow diagram, full HTML report

Interpreting I² and τ²

• I² < 25% — Low heterogeneity
• I² 25–75% — Moderate heterogeneity
• I² > 75% — High heterogeneity
• τ² — Between-study variance estimate

Fixed vs Random Effects

• Fixed-effect: Assumes all studies estimate the same true effect.
• Random-effects: Assumes the true effect varies between studies. More conservative; appropriate when heterogeneity is expected.

Bayesian Analysis

Uses the normal-normal conjugate model. The prior is Normal(μ₀, 1/τ₀) and is updated with the data likelihood. No MCMC needed — it's pure algebra. The Bayes Factor compares evidence for/against a non-zero effect.

PRISMA Guidelines

For reporting standards, see the PRISMA Statement.