MerQuR+Statistics Glossary

Statistics Glossary → Meta-regression

Meta-regressionconcept

A regression model used in meta-analysis to explain differences in effect sizes across studies by study-level characteristics (moderators).

When the effect sizes of the studies in a meta-analysis vary beyond what chance would produce (heterogeneity), the next question is where that variation comes from. In meta-regression the outcome is each study's effect size, and the predictors are study-level variables such as mean participant age, year of publication, study design or dose. Each study is weighted by its precision (inverse variance). Mixed-effects meta-regression also keeps in the model the between-study variance not explained by the moderators (τ²), and the share of heterogeneity explained by the moderators is reported with an R²-like measure.

Meta-regression is an observational analysis: because studies are not randomly assigned to moderator values, the relationships found cannot be read causally and may be confounded by other study features. Because study-level averages are used, there is a risk of drawing conclusions that do not hold for individuals (the ecological fallacy). With few studies, power is low and spurious findings arise easily; at least 10 studies per moderator are therefore recommended, and moderators should be specified in advance.

An everyday example

Thirty studies of the effect of exercise on depressive symptoms report very different results. A meta-regression with weekly exercise time as a moderator might show that the effect grows as exercise time increases. But if the longer programmes were also run with younger participants, it is impossible to tell whether the effect is due to duration or to age.

In MerQur

In Meta-Analysis, a mixed-effects meta-regression is fitted by choosing numeric or categorical moderators with the 'Meta-regression moderators' parameter.

Analyses where this concept appears

Meta-Analysis

Related concepts

Moderator variableSubgroup analysis