Case studies  /  Research

Random Effects Meta-analysis of Forty-Five Published Studies

Combining heterogeneous evidence into a defensible overall estimate, with publication bias investigated rather than assumed away.

ResearchSector
Random effects meta-analysis with meta-regressionMethod
R (metafor)Software
~45 studies, 20,000+ participantsScale
About six weeksDuration

The challenge

What the organisation came with

A research organisation needed to combine evidence across multiple published studies into a single robust estimate — the quantitative foundation for a body of work that would later be peer reviewed.

The studies differed substantially in methodology, participant characteristics and outcome measures. That between-study heterogeneity is not a nuisance to be minimised; it is a property of the evidence base, and a synthesis that ignores it produces a precise-looking estimate of something that does not exist.

Publication bias also had to be investigated. If studies finding no effect are less likely to be published, the published literature is a biased sample of the research conducted, and any synthesis of it inherits that bias.

The approach

How it was analysed, and why that method

Why random effects rather than fixed effects

A random effects model was used. It assumes the true effect varies between studies and estimates both the average effect and the extent of that variation, which is the honest representation of an evidence base drawn from different populations and designs.

Fixed-effects models were rejected because the assumption of a single common underlying effect was not supported by the data. Applying one would have produced artificially narrow confidence intervals — a more confident answer to a question the evidence could not support.

Explaining heterogeneity, and testing for bias

Meta-regression was used to examine whether study-level characteristics accounted for some of the between-study variation, which turns heterogeneity from a caveat into a finding. Publication bias was assessed using funnel plots and formal testing, and sensitivity analyses established how far the pooled estimate depended on any individual study.

Why the sensitivity analysis is not optional

A pooled estimate that changes materially when one study is removed is not a robust finding. Establishing that before publication is considerably better than a reviewer establishing it afterwards.

Delivered

What the client received

Meta-analysis report with pooled estimates and heterogeneity statistics
Forest plots and funnel plots
Meta-regression examining sources of heterogeneity
Sensitivity analyses
Reproducible R code

The value

What changed as a result

The findings supported evidence-based decision-making and formed the quantitative basis of a peer-reviewed publication.

Because the code was delivered with the report, the synthesis can be updated as new studies appear rather than being repeated from scratch.

About this case study

Written from the assigned statistician's own project notes. The client is not named and no identifying detail, data or figures are published. Scale and timeframe are approximate.

Send the data. Get a fixed quote.

Attach your dataset or just describe the project. A named statistician replies with a price and a deadline, usually within one working day.