Modelling dependent effect sizes in meta-analysis: Possibilities and limitations of three-level models
Year: 2010
Type: article
Abstract: Widespread meta-analytic techniques were developed to combine and compare independent effect size estimates. Nevertheless, meta-analysts are commonly confronted with dependencies. Effect sizes from different studies may for instance be statistically dependent if the same researcher or research group performed more than study. Within a study, dependent effect sizes may occur for instance if more than one sample was drawn, if several outcome variables are measured, or if several experimental groups are compared to a common control group. Several strategies for handling dependencies are discussed in the literature, such as using a multivariate approach, randomly choosing one measure per study or averaging effect sizes per study before combining effects over studies. One approach has received only very little attention although it is appealing since it is relatively easy to use and does not imply a loss of information: the use of three level meta-analytic models. In this presentation, we will discuss situations in which three-level models are or are not appropriate to deal with dependent effect sizes. The discussion will be illustrated using real data examples, and supported by the results of a simulation study.
Source: Lirias
Author: Wim Van Den Noortgate
Language: English
Subfield: Statistics, Probability and Uncertainty
Field: Decision Sciences
Domain: Social Sciences
Open Access status: green
OpenAlex ID: W2523088007
