Bayesian methods
Summary
Discover Bayesian statistics and learn how to combine data with prior knowledge to build efficient models. Gain hands-on experience with modern MCMC while working with your own data throughout the full analysis workflow. By the end, you will be able to independently develop and evaluate advanced statistical models for research and real-world applications.
About
Bayesian statistical analysis and frequentist statistical analysis are not the same. There are both philosophical and practical differences. The Bayesian approach formally integrates new evidence with pre-exisiting expertise.
The course starts with an introduction to the philosophical framework and the contrast between the Bayesian use of probability versus the traditional definition as a long-running frequency.
Subsequently the focus shifts to Bayesian computation and the Bayesian workflow, building from simple examples to Generalised Linear Mixed Models to illustrate the richness of Bayesian modelling. Specific attention is given to fitting regression models using Hamiltonian Markov Chain Monte Carlo, assessing convergence, and evaluating model fit. Practical experience is acquired by working on your own dataset and research question.
Prerequisites and selection
Entry requirements
The entry requirements of the course include a professional degree/ Bachelor's degree of at least 180 credits in health sciences, natural sciences, economics, or engineering. Further, R programming of at least 5 credits or equivalent, Introduction to biostatistics (STA110) or a course in mathematical statistics and/or probability theory of at least 9 credits or equivalent, English B/English 6 or equivalent, and Matematik 3b/3c or equivalent are required.
Selection
Selection is based upon the number of credits from previous university studies, maximum 165 credits.
After graduation
MCMC is a general approach that enables fitting both standard regression models with the typical choice of distributions and link-functions, and bespoke models. The software and workflow demonstrated in the course is readily adapted to a wider set of distributions, including hurdle-models and zero-inflated distributions.
This course does not cover alternative algorithms such as Integrated Nested Laplace Approximations, however, the general principle of model evaluation through posterior predictive checks applies.
This course does not cover adaptive trial designs, however, earning outcomes of the course provide essential knowledge and skills for studying Bayesian adaptive trials.
