The principal outcome of this project will be a new, formal framework for Bayesian cross-validation and model checking, assessment, and selection, which will help to build models with better predictive performance. Need for predictions is ubiquitous in science and society. The uncertainty quantification in predictions should be well calibrated, and we want to use the model making the most accurate predictions. We will develop methods that will detect smaller deficiencies and differences in models, which will reduce the risk of false scientific claims arising purely by chance when using finite noisy data. Users of our methods work on topics on a range of topics, including social science, health, sustainability, and climate change. This project will allow researchers in these fields and beyond to develop more elaborate models, better matching reality, which will serve to improve both scientific understanding and policymaking.

2026

2031