Bayesian Psychometric Network Modelling for Cross-sectional Data
My Ph.D. project is part of the ERC-funded Bayesian P-Nets project, headed by Maarten Marsman, which aims to develop new Bayesian methods for analyzing psychometric network models (i.e., undirected graphical models, also known as Markov random fields; Kindermann, & Snell, 1980). This subproject focuses on developing Bayesian edge
selection and structure averaging techniques for network models applied to cross-sectional psychological data (e.g., Madigan, & Raftery, 1994; Hoeting, Madigan, Raftery, & Volinsky, 1999). This approach accounts for the uncertainty of estimated parameters of any particular structure and the uncertainty of selecting the structure itself, yielding a novel framework for exploratory and confirmatory psychometric network modeling.
This Ph.D. project aims to extend the work proposed by Marsman, Huth, Waldorp, and Ntzoufras (2020) for the Ising model (Ising, 1925), a Markov Random Field model for binary data; developing new structure averaging techniques, and new models for mixed variable types (e.g., binary, ordinal, and continuous data).
The Ph.D. project comprises four subprojects:
- Develop prior specifications for a network’s parameters and architecture in psychological contexts;
- Explore and develop network models that cover different types of cross-sectional data (e.g., binary, ordinal, and continuous) and estimation methods for their Bayesian analysis.
- Explore and develop structure averaging techniques that overcome the difficulties of model-averaged assessment (e.g., Occam’s window and stochastic search methods);
- Use the new methods to reanalyze available psychological data to determine the statistical evidence of published findings. With the above in mind, I believe that my Ph.D. project fits the IOPS graduate school since it entails the development of new statistical techniques for analyzing psychological data.
Prof. dr. H.L.J. (Han) van der Maas
Dr. M. (Maarten) Marsman
The European Research Council
1 September 2022 – 1 September 2026