Comparability-Based Informative Priors for Bayesian Replication in Structural Equation Modeling (CIBR-SEM)

• Applicants: Dr. Aron Fink; Dr. Christoph König
• Subject Area: Personality Psychology, Clinical and Medical Psychology, Methodology
• Term: since 2026
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 570467746

Bayesian data analysis offers a different perspective on replicability of scientific results, a topic that received widespread attention over the last decade. From this perspective, as more evidence on a given research question emerges, an inherent iterative updating process, implemented by the use of informative prior distributions, reduces the uncertainty regarding the existence or absence of a given effect. Such an approach is especially fruitful for psychological research, where most studies are conceptual, rather than direct replications. The prevalence of conceptual replications, and the associated heterogeneity in study characteristics and outcomes, however, has been frequently cited as one of the primary reasons for the failure of a large proportion of studies to replicate effects without conclusive support. At the same time, it is also a major obstacle for an increased use of informative prior distributions in Bayesian data analyses. Due to intended or unintended differences in sampled populations, utilized instruments, and diverging outcomes, it remains unclear which studies can actually be used to specify informative prior distributions for conceptual replications. This uncertainty often leads to the neglect of these distributions. Specifically, in multivariate methods such as structural equation modelling (SEM), informative prior distributions are seldom utilized. To address this issue, CIBR-SEM offers substantial advancements in two key areas relevant for psychological research and beyond: a) the systematic development and application of informative prior distributions, and b) an extension to existing methods for assessing replication success in a Bayesian context. Specifically, CIBR-SEM refines an innovative combination of state of the art methods for propensity score determination and meta-analytic structural equation modelling and develops a novel procedure to quantitatively integrate different sources of between-study heterogeneity for the specification of informative prior distributions in the context of Bayesian SEM. CIBR-SEM will illustrate how the resulting comparability measure can be used to account for multiple sources of variation when conceptualizing or integrating conceptual replications and testing their success. In a series of comprehensive simulation studies and real-data applications, the procedure will be tested and evaluated as a Bayesian method to assess replication (success) of Bayesian SEM. Furthermore, the new methodological procedures will be embedded in a comprehensive dissemination strategy including, among others, the development of an R-package and a project specific workshop, aiming to facilitate their integration into psychological research and beyond. In sum, CIBR-SEM will provide a conceptual and methodological advancement that is not only relevant and adaptable to psychological research but also holds foundational scientific relevance for general Bayesian replication research.

• DFG Programme: Research Grants