Project List

Bayesian Structural Equation Modeling (BSEM) has gained significant attention due to its ability to resolve common issues found in frequentist approaches, such as nonconvergence, Heywood cases, sample size limitations, and inadmissible solutions. Furthermore, BSEM can estimate complex models that classical maximum likelihood methods often struggle with. A crucial component of BSEM is the incorporation of prior knowledge via prior distributions, which provides a unique advantage over frequentist methods by allowing previously known information to be transparently included in model specifications. Proper prior elicitation is essential for translating domain expertise into probability distributions, thus improving the accuracy and reliability of the model. Despite this, the development and widespread adoption of robust prior elicitation techniques in BSEM remain limited. This project aims to advance the field of BSEM through innovative computational methods and practical applications, with a focus on GPU processing and its potential to enhance the efficiency of Bayesian computations. By leveraging parallel computing capabilities provided by modern graphic processing units, the project seeks to significantly accelerate the computational processes involved in BSEM. Additionally, the project will explore the use of BSEM in developing psychometric instruments, employing techniques such as Markov Chain Monte Carlo (MCMC) using the No-U-Turn Sampler (NUTS) for latent model estimation. These innovations promise to offer deeper insights into measurement precision and enhance the validity of psychometric instruments across diverse populations. The expected outcomes of this project include improved computational efficiency, optimized model estimations, and broader adoption of Bayesian methods in applied research settings. By addressing both theoretical and practical challenges, this project aims to contribute to the broader use of BSEM in psychometric analysis.

Despite decades of devoted research, cancer remains a tremendous health threat and societal burden. Europe's Beating Cancer Plan aims to improve the lives of more than 3 million people by 2030 by improving prevention, early detection, diagnostics, therapeutics, and quality of life. The biggest single hurdle here is the highly inadequate way cancer data, both from research and healthcare, are still being dealt with. While other areas of society (e.g. e-finance, e-commerce, logistics, travel, meteorology, etc.) have fully exploited advances in data and information technology to serve organisations as well as individual consumers, so far this has failed in the health domain. Consequently, cancer data are hard to Find, Access, make Interoperable and Reuse. Evidently this is not caused by lack of suitable technology, but rather by organisational, social and cultural causes. Inherently, solving the problem requires a cultural shift from the current craftsmanship approach to cancer research and data, to a drastic collaboration model at industrial scale. CANDLE therefore aims to scale-up and improve existing (inter)national health data infrastructures, align maximally with national EHDS implementations in member states, including HDAB’s, DAAMS’s and SPE’s. CANDLE will also identify and resolve potential barriers (https://www.health-ri.nl/en/participation/obstacles-removal-trajectory ) that jeopardize effective implementation of UNCAN.eu and ECPDC digital platforms. CANDLE aims to equipe data users and NCDN developers with a ‘ready-to-use' CANDLE Resource Kit in a process oriented (research journey, patient journey, data life cycle) way. In summary, CANDLE will provide an avenue towards a successful and highly desired data transformation in European cancer research and serve as a catalyzer for the UNCAN.eu and ECPDC platforms by advancing the development of NCDNs to reach the goal of the Cancer Mission and Europe’s Beating Cancer Plan, i.e. reducing the burden of cancer.

The already established consortium of Aston University (UK), University College London (UK), Ruralis University (Norway), University of Turin (Italy), and University Institute of Lisbon (Portugal) will seek funding from the EU's ‘100 Climate-Neutral and Smart Cities by 2030’ initiative. Our proposal includes developing a scalable tool to evaluate and guide urban mobility policies, supporting sustainable development and the 15-minute city concept, especially post-COVID-19. This tool will analyze mobility's complex nature, integrating environmental sustainability and regional urban system characteristics like geography and demographics. AIenhanced, it will assess impacts on affordability and accessibility, providing localized insights. We seek pumpprimingfunds for two residential workshops at Aston University to solidify collaboration, complete application writing, and establish a proof of concept

Bayesian structural equation modeling (BSEM) is receiving increasing interest primarily due to its ability to address some of the issues found in the mainstream frequentist approach (e.g., nonconvergence, Heywood cases, sample size limitations, and inadmissible solutions). Additionally, BSEM allows for the fitting of complex models that classical maximum likelihood methods might struggle to handle. A critical component of any Bayesian analysis is the prior distribution of the unknown model parameters. A key distinction between Bayesian structural equation modeling and frequentist structural equation modeling is the use of priors. Researchers may be skeptical about the subjectivity of prior distributions and their impact on Bayesian modeling. However, priors are a significant advantage of using Bayesian statistics, as they allow previously known information to be transparently and directly included in the model specification. Proper prior elicitation is essential for translating knowledge and judgment about a phenomenon into a probability distribution. Priors allow for the quantification of uncertainty and encapsulate available knowledge about the parameters before observing the data. There are several ways to translate prior knowledge into distribution parameters. Results show that researchers tend to rely on weakly informative priors (i.e., small-variance priors). However, prior elicitation in Bayesian structural equation modeling still has a long way to go in terms of development and widespread adoption.

Traditional psychometric approaches often rely on frequentist models, which can be limited in handling uncertainty and incorporating prior knowledge. By adopting a Bayesian framework, this project aims to enhance psychometric analysis through more flexible, probabilistic modeling techniques that allow for dynamic updates as new data becomes available. The project focuses on the creation and adaptation of psychometric instruments for diverse populations. Advanced computational techniques, including Markov Chain Monte Carlo (MCMC) using the No-U-Turn Sampler (NUTS), are employed for latent model estimation. The Bayesian approach provides a deeper understanding of measurement precision and contributes to the development of psychometric instruments with strong validity evidence.