In meta-analyses, results of multiple studies that studied more or less the same research question are combined and compared. Nowadays, meta-analysts typically use random effects meta-analytic models, which are models that take into account that observed effects from multiple studies may not only differ from each other due to sampling variance, but also due to systematic differences between studies in the setup, the population sampled from, the process and so on. Pioneering work on random effects meta-analytic models was done by Raudenbush and Bryk (1985, 1992), who showed that such a random effects model can be considered as special cases of multilevel models. An advantage of the use of the multilevel modeling framework for meta-analysis it its flexibility. For instance, multilevel models can include additional random effects. Still, in a systematic review, we found that before 2014, only a few meta-analyses with more than one random effect were published, whereas since then the number is still limited but increasing fast.

The current PhD-project has two aims. First, only very little methodological research has been done on the use of multilevel models for meta-analysis. We want to obtain a better understanding of the performance of multilevel models in comparison to other meta-analytic approaches for the meta-analysis of dependent effect sizes. Second, we want to contribute from a multilevel modeling perspective to the development of specific techniques and models (especially for network meta-analysis), especially in scenarios in which various dependencies are involved. Both the meta-analysis of group experimental studies and the meta-analysis of single-case experimental studies are the focus of the current study.