Explainer · Plain-language

What Is a Meta-Analysis? Methods & How to Read One | CASRAI

A meta-analysis is a statistical technique that pools quantitative results from multiple independent studies on the same research question to produce a single, more precise estimate of an effect. It is the highest-ranked form of primary synthesis in the evidence hierarchy.

CASRAI plain-language explainers — clear answers to recurring research-administration questions

The step most authors miss

Doing CRediT right? Don’t stop at the statement.

A CRediT statement credits you inside one paper. The recognition CRediT was built for happens when those roles are tied to you, persistently. Sign in with your ORCID — free — and claim your CRediT contributions on casrai.org, the home of the standard. They become a verified, portable part of your identity, not a line that disappears into one PDF.

Free: claim your contributions, then export a journal-ready CRediT statement, schema.org structured data, JATS XML, CSV or BibTeX — and preview your public profile. A membership publishes that profile publicly and verifies the journals you serve.

Origins and what a meta-analysis produces

The term "meta-analysis" was coined by Gene Glass in 1976 to describe the statistical integration of results from a collection of studies. The primary output is a pooled effect size — a single summary statistic (such as Cohen’s d for standardised mean difference, an odds ratio, relative risk, or mean difference) with a confidence interval — that combines the estimates from all included studies weighted by their precision. This pooled estimate is typically displayed in a forest plot, where each horizontal line represents one study and a diamond at the bottom shows the pooled result.

Fixed effects, random effects and heterogeneity

Two statistical models underpin meta-analysis. A fixed-effects model assumes all studies are estimating the same underlying true effect; a random-effects model (DerSimonian–Laird) assumes that the true effect varies between studies — which is usually more realistic. Heterogeneity (variability in true effects across studies) is quantified by the I² statistic: values below 25% indicate low heterogeneity, 25–75% moderate, and above 75% high. The Q test provides a significance test for heterogeneity. Unexplained high heterogeneity is a reason to interpret pooled estimates cautiously and to investigate moderating variables in a meta-regression.

Publication bias and how to detect it

Publication bias arises because studies with statistically significant results are more likely to be published than null results, distorting the pool of studies available for meta-analysis. It is assessed visually through a funnel plot (a scatter of study effect sizes against standard error — asymmetry suggests bias) and statistically using Egger’s test or Begg’s rank correlation test. The trim-and-fill method estimates how many studies are "missing" and adjusts the pooled estimate. Despite these tools, publication bias is hard to eliminate and remains a major limitation of the technique.

PRISMA and Cochrane Reviews

Meta-analyses are reported according to the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) statement, updated in 2020 to include network meta-analyses and other extensions. PRISMA specifies a flow diagram documenting the number of records identified, screened, assessed for eligibility, and included. The Cochrane Collaboration, founded by Iain Chalmers in 1992, produces the world’s most authoritative health-care systematic reviews and meta-analyses, published in the Cochrane Database of Systematic Reviews, and maintains the RevMan software for meta-analysis.

Key facts

At a glance

Definition: Statistical technique pooling effect sizes across multiple studies
Coined by: Gene Glass (1976)
Key output: Pooled effect size with confidence interval, displayed in a forest plot
Models: Fixed-effects vs random-effects (DerSimonian–Laird)
Heterogeneity: I² statistic: <25% low, 25–75% moderate, >75% high
Bias detection: Funnel plot, Egger's test, Begg's test, trim-and-fill
Reporting: PRISMA 2020 statement; Cochrane Reviews as gold standard

Common misconceptions

What people often get wrong

Often heard: A meta-analysis can be done independently of a systematic review.

Actually: In principle yes, but best practice embeds meta-analysis within a systematic review so that the pool of studies is identified through a comprehensive, reproducible search rather than an ad-hoc selection that risks bias.

Often heard: A large I² always means the meta-analysis result is unreliable.

Actually: High heterogeneity is a signal to investigate sources of variability (via subgroup analysis or meta-regression), not necessarily to discard the analysis. The pooled estimate may still be meaningful if heterogeneity can be explained.

Often heard: More studies always make a meta-analysis better.

Actually: Quality matters more than quantity. Including poorly designed or biased studies degrades the pooled estimate; the principle "garbage in, garbage out" fully applies to meta-analysis.

Going deeper