Project
Bayesian Penalisation and Variable Selection in Relational Event Model

This research advances psychometric methodology by (1) introducing Bayesian shrinkage priors for parameter regularization, (2) developing efficient computational techniques for estimating penalized REMs, and (3) comparing Bayesian penalization approaches to existing frequentist methods, such as LASSO or stepwise selection. These contributions will provide researchers with more reliable tools for modeling relational event data while addressing key issues like multicollinearity and parameter sparsity.

Beyond methodological advancements, this project has significant implications for social and behavioral sciences. By improving the interpretability and predictive power of REMs, the developed Bayesian framework will enable researchers to uncover more precise insights into the temporal dynamics of social interactions. The ability to model evolving relational structures with greater accuracy can benefit various fields, including psychology, sociology, and network science, where understanding patterns of communication, collaboration, and influence is crucial. Additionally, the open-source R package developed in this project will make these advanced methods widely accessible, facilitating their application in empirical research. By developing and evaluating novel Bayesian methods, this project directly aligns with IOPS’s goals of advancing psychometric and sociometric techniques which can be used in psychology and social network analysis.

Supervisors
Prof. Ir. Mirjam Moerbeek
Dr. Sara van Erp
Dr. Mahdi Shafiee Kamalabad

Financed by
Private budget

Period
1 December 2024 – 30 November 2028