Using collaboration networks to identify authorship dependence in meta-analysis results.

Meta-analytic methods are powerful resources to summarize the existing evidence concerning a given research question and are widely used in many academic fields. Meta-analyzes can also be used to study sources of heterogeneity and bias among results, which should be considered to avoid inaccuracies. Many of these sources can be related to study authorship, as both methodological heterogeneity and researcher bias may lead to deviations in results between different research groups. In this work, we describe a method to objectively attribute study authorship within a given meta-analysis to different research groups by using graph cluster analysis of collaboration networks. We then provide empirical examples of how the research group of origin can impact effect size in distinct types of meta-analyzes, demonstrating how non-independence between within-group results can bias effect size estimates if uncorrected. Finally, we show that multilevel random-effects models using research group as a level of analysis can be a simple tool for correcting for authorship dependence in results.