Short Courses in Applied Statistics

Course outline

Bayesian statistical methods utilise prior information about model parameters in the inference process. Although the idea is not new, it is only relatively recently that modern computational methods have made Bayesian data analysis a practical possibility. In particular, simulation methods such as Markov chain Monte Carlo (MCMC), implemented in the WinBUGS software, enable the Bayesian analysis of an exceptionally wide range of statistical models. Even when no prior information is available, the flexibility offered by this approach far exceeds that of any other modelling framework.

The emphasis in this course is on practical data analysis, although the essential theory will be explained. Starting from simple single-parameter models, the course will develop an approach to the analysis of quite complex data structures. The course will include a practical introduction to WinBUGS and will also make use of the R package. Participants will have the opportunity to use WinBUGS in the practical sessions.

Who should attend?

Statisticians and data analysts who wish to use a Bayesian approach in analysing their data. Even those who are not comfortable with using prior information in their analysis will find the flexible modelling made possible by MCMC methods a powerful tool. No prior knowledge of WinBUGS or R will be assumed.

How you will benefit

You will extend your data analysis skills to cover a very wide class of modelling, including the use of prior information. You will learn how to use specialised software for Bayesian data analysis.

Course content

Likelihood, prior and posterior distributions and the use of Bayes' theorem
Bayesian analysis of single-parameter models and simple multi-parameter models
Conjugate, non-informative and informative priors
Simulation of posterior distributions; posterior summaries
MCMC, Gibbs sampling and Metropolis-Hastings; MCMC diagnostics
Linear models, generalised linear models and model selection
Model checking
Hierarchical Bayesian models
Using WinBUGS