ABSTRACT
All (random-effects model) meta-analyses are based on a multilevel model. When a third layer is added to this multilevel model, one can speak of a three-level model. Three-level models are particularly well suited to handle clustered data with dependent effect sizes.
This chapter begins with an explanation of the “multilevel” nature of meta-analysis, and how a conventional meta-analysis can be extended using a three-level model. A description of the two heterogeneity variance components in three-level meta-analysis models is provided.
The chapter also includes a hands-on tutorial on how to fit nested three-level models in R, based on a real-life data set. Lastly, the extension to a three-level mixed-effects models with categorical or continuous predictors is described, and illustrated using an example in R.
