Random-Effects vs Fixed-Effect Meta-Analysis: Which Should You Use?

The Core Difference

The fixed-effect model assumes all studies in the meta-analysis share one true underlying effect size, and that observed differences between studies are due to sampling error alone. The random-effects model assumes that the true effect size varies across studies — that each study estimates a slightly different true effect drawn from a distribution of effects.

When to Use Fixed-Effect

Fixed-effect is appropriate when: (1) all studies are very similar in population, intervention, and outcome; (2) you are specifically interested in the studies at hand rather than a broader population; (3) the studies are considered exchangeable — essentially replicates of each other.

In practice, fixed-effect is rarely justified for clinical research, because studies almost always differ in some meaningful way.

When to Use Random-Effects

Random-effects is the default for most clinical meta-analyses. It produces wider confidence intervals than fixed-effect when heterogeneity is present, which is more honest about uncertainty.

Tau-Squared Estimators

DerSimonian-Laird (DL). The most widely used estimator. Fast and straightforward, but can underestimate τ² when k is small.

Paule-Mandel (PM). Generally preferred for small meta-analyses (k < 10). More accurate τ² estimates, especially with few heterogeneous studies. Verflux supports both.

Heterogeneity Thresholds

I² of 0–25%: probably low heterogeneity. 25–50%: moderate. 50–75%: substantial. 75–100%: considerable. These are rough benchmarks — always consider the magnitude of τ² and the clinical context.