Common-effect meta-analysis
A common-effect meta-analysis combines study estimates under the assumption that they estimate the same true effect.
Fit
When to use it, and when not to
Use it when
- Studies address a sufficiently similar question and a shared true effect is scientifically credible.
- You have comparable effect estimates and standard errors from independent studies.
Not when
- You expect true effects to vary across settings. Consider a random-effects meta-analysis. Read: Random-effects meta-analysis and prediction
- The studies answer incompatible questions. Use a structured synthesis without pooling. Read: Systematic review
- You jointly summarize sensitivity and specificity. Use a diagnostic accuracy meta-analysis. Read: Meta-analysis of diagnostic test accuracy
Before you start
What you need
- A common effect scaleAlign treatment direction, outcome definitions and follow-up. Pool ratios on their log scale.
- Each study’s estimate and standard errorFor inverse-variance pooling, square each standard error to obtain the within-study variance.
- The reason for assuming one effectChoose the model before seeing which result is statistically significant.
- The studies estimate one true effect.How to check: Compare populations, treatment delivery and outcome timing; a heterogeneity test cannot prove the assumption.
- Study estimates do not count the same people twice.How to check: Check overlapping cohorts and shared controls. Model dependence when several estimates come from one study.
- Pooling does not remove study bias or selective reporting.How to check: Assess risk of bias and missing results. Prespecify sensitivity analyses for fragile inputs.
Plain language
How it works
For ordinary inverse-variance pooling, give each estimate weight equal to one divided by its variance. Divide the weighted sum by the total weight. The pooled standard error is the square root of one divided by total weight. Whitehead and Whitehead set out a general framework using study estimates and their uncertainty.
A normal confidence interval adds and subtracts 1.96 pooled standard errors. This approximation needs suitable study estimates; sparse binary data may need a different implementation. DerSimonian and Laird contrast shared-effect pooling with models allowing variation between studies. Report model, effect scale, interval method, study weights and risk of bias.
Worked example
A worked example
- Setting
- Two invented studies estimate treatment minus control hospital stay in days.
- Question
- What is the shared mean difference under a common-effect assumption?
- Approach
- Study estimates are -2 and -4 days, with standard errors of 1 and 2 days. Inverse-variance weights are 1 and 0.25.
- Result
- The normalized weights are 80 and 20 in 100. The pooled mean difference is -2.40 days; its standard error is 0.894 days. The normal 95% confidence interval is -4.15 to -0.65 days.
- What it means
- If control stay averages 10 days, the assumed shared difference corresponds to 7.6 days with treatment. This interval does not describe variation across settings.
In practice
Recent studies PubMed search · not vetted
Watch out
Common mistakes
The one to remember
A common-effect interval describes uncertainty about one shared effect under that assumption. A nonsignificant heterogeneity test does not establish that the assumption is credible.
- Selecting common effect because the heterogeneity test is nonsignificant.Instead: Judge the shared-effect assumption from the question and studies, with uncertainty about heterogeneity.
- Averaging ratios on their original scale.Instead: Use a suitable common scale, such as log odds ratios, and transform the result back.
- Counting several reports of one trial as separate studies.Instead: Identify unique studies and handle correlated estimates explicitly.
Inside Mallard
How Mallard uses this
Mallard includes this article in Learn search and the Atlas. The plan mapper does not link to this article yet.
analysis.common_effect_meta_analysis, the id a plan cue records.Related Learn pages
Related designs and methods
Evidence
Sources & further reading
- A general parametric approach to the meta-analysis of randomized clinical trials (opens in a new tab) · Stat Med · doi 10.1002/sim.4780101105
- Meta-analysis in clinical trials (opens in a new tab) · Control Clin Trials · doi 10.1016/0197-2456(86)90046-2