• Statistical Concepts
• WinBUGS
  • WinBUGS demonstrations
• The `Classic' BUGS software
  • Classic BUGS demonstration
• CODA
• Keeping in touch

BUGS is a program that carries out Bayesian inference on complex statistical problems for which there is no exact analytic solution, and for which even standard approximation techniques have difficulties. Conditional independence assumptions mean that it is often convenient to represent the essential structure of the problem as a graphical model. A Markov chain Monte Carlo (MCMC) approach to numerical integration is used: BUGS Version 0.6 generally uses univariate Gibbs sampling, although can use a simple Metropolis-within-Gibbs routine when necessary: WinBUGS has a more sophisticated univariate Metropolis sampler.

Typical applications include generalised linear mixed models with hierarchical, crossed, spatial or temporal random effects; latent variable models; frailty models; measurement errors in responses and covariates; censored data; constrained estimation; missing data problems, and any analysis in which informative prior information needs to be incorporated.

BUGS provides a declarative language for straightforward specification of statistical models based on the assumed graphical structure, although there are some restrictions on the class of models that can be currently be analysed. A compiler then processes the model and data and sets up the sampling distributions required for the Gibbs sampling. Finally, appropriate sampling algorithms are implemented to simulate values of the unknown quantities in the model. Work on BUGS began in 1989, and led to the release of what we now call `Classic' BUGS for a number of platforms. Recent development has focussed on WinBUGS .

The most advanced version of BUGS is now WinBUGS running under Windows. This comprises a program for graphical specification of models called DoodleBUGS , and an educational version of WinBUGS . This allows analysis of all the examples available for classic BUGS , and for users to try out small examples of their own. An unrestricted version of WinBUGS 1.3 is currently freely available for users who register.

• Watch WinBUGS - The Movie! for a step-by-step introduction to a very simple WinBUGS session. (large file, requires Flash)
• Click here for a (very old and grubby) screendump of WinBUGS in action, doing the hepatitis measurement error problem described in Gilks, Richardson and Spiegelhalter (1996), Chapter 2.
• Saghir Bashir has also produced a set of tutorial slides on using Doodles and running WinBUGS 1.3, which can be downloaded as a pdf file.
• Apart from the examples provided in the program, there are many worked examples and explanatory papers that other researchers have kindly made available on the Web - see the BUGS on the web page.

The main documentation is for Version 0.5, which is still available on this web site. Version 0.6 has additional facilities and is also available, with an addendum for the manual.

BUGS currently provides only limited facilities for checking convergence, summarising the accumulated samples and performing model diagnostics - these procedures are best handled externally to the program. A menu-driven set of S-Plus functions (CODA) are supplied with BUGS to calculate convergence diagnostics and graphical and statistical summaries of the simulated samples if desired (although it is not necessary to have S-Plus in order to run BUGS ).

• A `walkthrough' of a Classic BUGS analysis is available here.

This set of S-plus functions contains a wide range of diagnostics and plots for presenting the results of an MCMC analysis and checking convergence of the simulation. Version 0.4 of CODA is now available, but does not work for S-plus Version 5.0 and above. However, a fully functioning version of CODA using the free R software has been developed by Martyn Plummer, and the program BOA is also available from Brian Smith. See the relevant README file for more details.

We are keen to get feedback from all users - there are two ways of getting in touch;

• Mail to bugs at mrc-bsu.cam.ac.uk for specific questions about getting hold of the software, any difficulties in installation, and any crashes and difficulties with error messages. We shall try to reply as soon as possible.
• We have also set up a discussion list as a forum for sharing ideas and asking questions about modelling issues using BUGS. It is NOT intended for problems with crashes, error messages, or `my program won't work'! - please send such queries to bugs at mrc-bsu.cam.ac.uk. Details of how the list works can be found here.