In meta-analysis, which statement best distinguishes a fixed-effects model from a random-effects model?

In meta-analysis, which statement best distinguishes a fixed-effects model from a random-effects model?

• A. Fixed-effects: true effect is the same across studies; Random-effects: true effects vary between studies
• B. Fixed-effects: studies are different; Random-effects: studies are identical
• C. Fixed-effects uses IPD; Random-effects uses aggregated data
• D. Fixed-effects requires more studies; Random-effects requires fewer

Answer: (A)

In meta-analysis, the key idea is how to handle differences in effect sizes across studies. In a fixed-effects model, we assume there is one true effect size that applies to all included studies, and any differences observed are due to sampling error within each study. In a random-effects model, we allow the true effect size to differ from study to study, treating these study-specific effects as draws from a distribution (often assumed normal). This difference changes both interpretation and weighting: fixed-effects weights studies by their within-study variance and attributes all variability to sampling error, while random-effects adds between-study variance to the weighting, typically making the pooled estimate more conservative and widening confidence intervals. The statement that best captures this distinction is the one that says the true effect is the same across studies under fixed-effects and that true effects vary between studies under random-effects. The other options misstate the core idea: the design differences of studies or the data types used (IPD vs aggregated) aren’t what define the models, and the number of studies required doesn’t determine whether a fixed- or random-effects approach is inappropriate or appropriate.

In meta-analysis, the key idea is how to handle differences in effect sizes across studies. In a fixed-effects model, we assume there is one true effect size that applies to all included studies, and any differences observed are due to sampling error within each study. In a random-effects model, we allow the true effect size to differ from study to study, treating these study-specific effects as draws from a distribution (often assumed normal). This difference changes both interpretation and weighting: fixed-effects weights studies by their within-study variance and attributes all variability to sampling error, while random-effects adds between-study variance to the weighting, typically making the pooled estimate more conservative and widening confidence intervals.

The statement that best captures this distinction is the one that says the true effect is the same across studies under fixed-effects and that true effects vary between studies under random-effects. The other options misstate the core idea: the design differences of studies or the data types used (IPD vs aggregated) aren’t what define the models, and the number of studies required doesn’t determine whether a fixed- or random-effects approach is inappropriate or appropriate.