Dissertation
The integration of meta-analysis and classification & regression trees: meta-CART
In meta-analysis, heterogeneity often exists between studies. In such cases, it is essential to investigate the sources of heterogeneity and understand the relationship between effect size and study characteristics (i.e., moderators).
- Author
- Li, X.
- Date
- 27 February 2020
- Links
- Thesis in Leiden Repository
In meta-analysis, heterogeneity often exists between studies. In such cases, it is essential to investigate the sources of heterogeneity and understand the relationship between effect size and study characteristics (i.e., moderators). Applying tree-based methods in meta-analysis is a promising alternative for conventional meta-regression, since trees excel at modeling interactions and non-linear relationships and provide easily interpretable results. In this thesis, we propose a method called meta-CART, which integratesclassification and regression trees (CART) into the framework of meta-analysis. This method identifies subgroups of homogeneous studies by searching influential moderators that can explain the heterogeneity, and performs subgroup analysis to test the significance of the identified moderators and estimate the subgroup effect sizes.