Fixed-Effect vs. Random-Effects Meta-Analysis

Choosing between a fixed-effect and random-effects model is one of the first decisions in running a meta-analysis, and it changes both the pooled estimate and how it should be interpreted.

Fixed-effect model

Assumes every included study is estimating the exact same true effect, and any observed differences between studies are due purely to random sampling error. Weights are based mostly on sample size — larger studies dominate the pooled result.

Random-effects model

Assumes the true effect varies across studies — because of differences in populations, settings, or intervention delivery — and estimates the average of that distribution of effects. Weighting is more balanced across studies, since it accounts for between-study variance (tau²) as well as within-study error.

How to choose

In practice, random-effects is the more common and defensible default in the social and health sciences, because true clinical and methodological diversity between studies is the norm rather than the exception. Fixed-effect is appropriate only when you have strong reason to believe all studies estimate one true effect — for example, tightly controlled replications of the same trial protocol.

Check your heterogeneity (I²) results either way — substantial heterogeneity under a fixed-effect assumption is a sign the model doesn't fit your data.

Why this decision matters

The two models can produce meaningfully different pooled estimates and, more importantly, different confidence interval widths — random-effects models typically produce wider, more conservative intervals. Choosing the wrong one, or not justifying the choice, is a common point of reviewer pushback.