From Generative Network Models to Statistical Evidence: A Simulation Study of Bayesian Evidence Synthesis and Hierarchical Modelling
Keywords
Loading...
Authors
Issue Date
2025-09-23
Language
en
Document type
Journal Title
Journal ISSN
Volume Title
Publisher
Title
ISSN
Volume
Issue
Startpage
Endpage
DOI
Abstract
Abstract
This thesis compares the performance of Bayesian Evidence Synthesis (BES) and Bayesian Hierarchical
Modelling (BHM). BES is a novel method for performing meta-analysis when traditional
meta-analysis might not be feasible. This can be useful in settings such as generative network
modelling, where transforming effect sizes onto a common scale is not always possible. Due to
BES aggregating evidence from individual studies, it does not require such transformations.
However, it also does not model between-study variance, whereas BHM does. This study, therefore,
examines how Bayesian Evidence Synthesis behaves in comparison to BHM, using a linear
regression model under differing levels of: between-study heterogeneity, correlation, effect size
differences, and the number of studies. It was found that increased heterogeneity can lead BES
to give greater support to either the correct or the incorrect hypothesis, depending on the conditions,
and that introducing correlation may amplify this effect even further. Whereas correlation
by itself seems to affect both models similarly. The findings highlight both the potential and the
limitations of BES as an alternative to traditional meta-analysis, particularly in research domains
such as generative network modelling. Recommendations and directions for future work
are provided, including the need for frameworks that can apply BES in more practical GNM
settings.
Description
Citation
Supervisor
Faculty
Faculteit der Sociale Wetenschappen