Introduction to Bayesian Inference: Practical Exercises You will be using WinBUGS 1.4.3 for these practicals All the data and other ﬁles you will need for the practicals are provided in a zip. And in the on-line WinBUGS User Manual (see the Help menu in WinBUGS). PracticalExercises 2. This should work for any version of Windows, for 32 or 64 bit. It includes the key for unrestricted use, and has been updated to version 1.4.3. Easier installation (32 bit Windows only) An easier way to install WinBUGS that only works on 32 bit machines is to download and install WinBUGS14.exe. Then download and install the patch for 1.4.3. WinBUGS User Manual: Version 1.4. Models were implemented in WINBUGS software, version 1.4 (Spiegelhalter et al., 2003) using Markov chain Monte Carlo (MCMC) algorithms to generate posterior.

Course dates

29 - 31 January 2018.

Course duration

3 days (approximately 18 hours teaching).
Registration will start at 9am on the first day, the course will finish by 4.30pm on the final day.

Course tutors

Professor Tony Ades, Dr Sofia Dias, Professor Nicky Welton(course organisers) Dr Hayley Jones, David Phillippo and others.

Course aims

This course aims to provide an introduction to Bayesian data analysis applied to healthcare decision-making through examples that showcase the potential of the methods and their implementation in WinBUGS.

Course objectives

By the end of the course participants should:

understand the basic principles of Bayesian data analysis;
be able to carry out elementary data analysis in WinBUGS;
have some insight into the circumstances in which there are advantages in a Bayesian approach.

Who the course is intended for

This course is designed for statisticians and those interested in quantitative methods for decision making, including health economists.

Prerequisites

Winbugs 1.4.3

This course is at an intermediate or higher level. Participants should have a sound knowledge of statistical methods, including multiple regression, generalised linear models and hierarchical models.

The course will be built around analyses carried out with WinBUGS software. Participants will be expected to have downloaded WinBUGS 1.4.3, read through the Tutorial in the User Manual, and tried out some of the examples before coming on the course.

WinBUGS version 1.4.3 will be available on the machines in the teaching room. Participants who are familiar with OpenBUGS will not experience any difficulties with WinBUGS 1.4.3.

Essential pre-course preparation:

Participants who have not used WinBUGS 1.4.3 before should download the software, including the patch and key (see instructions below), take the Tutorial in the User Manual (under Help -> User manual), and try out at least the BLOCKER example (under Help -> Examples Vol I).
WinBUGS 1.4.3 including the patch and key can be downloaded from http://www.mrc-bsu.cam.ac.uk/software/bugs/. The best way to install it on your machine is to download a zipped version of the whole file structure and unzip it into Program Files or wherever you want it. This includes the key for unrestricted use. Then, download and install the patch for version 1.4.3. To check that you have the patched version, open WinBUGS, go to the “Help” menu and click “About WinBUGS”. This should show the current version as 1.4.3. The software, patch and key are free.

For those using a Mac: To run WinBUGS on a Mac you will first need to install a Windows emulator (e.g. WINE). You can then install WinBUGS on that and it should work. For more precise instructions please check for advice online.

Course outline

Introduction to Bayesian methods.
Basic principles of Markov chain Monte Carlo.
Running WinBUGS: initial values, burn-in, posterior simulation, use of the menus and graphics.
Series of worked examples covering various topics, including:
- hierarchical models;
- analysis of sensitivity, specificity and prevalence;
- predictive cross-validation;
- interactions in a factorial design trial;
- error in covariates regression;
- multiple-parameter evidence synthesis;
- diagnostics tests with no gold standard;
- missing data / bias in meta-analysis;
- Bayesian decision analysis;
- Expected Value of information.
Round-up and Summary: advantages of a Bayesian approach.
Open discussion and feedback.

The course will comprise about 18 hours of teaching time over 3 days in an informal atmosphere. Teaching will consist of lectures and WinBUGS practicals.

Course tutors will be available throughout to help participants with the practical exercises. Participants are welcome to bring along their own data analysis problems and discuss with tutors whether a Bayesian approach would be appropriate.

Booking

Bookings will be accepted from 10th October.

Course fee

£660

More information on course fees, fee waivers and reduced prices.

Course venue

Bristol Medical School
Canynge Hall
39 Whatley Road
Bristol
BS8 2PS
United Kingdom

Course refreshments

We provide morning and afternoon refreshment breaks, including tea and coffee, biscuits and fresh fruit.

If you have specific dietary needs we ask that you let us know in advance.

Lunch is not included. There are a range of local cafes and supermarkets nearby for students to purchase lunch.

Accommodation

Information about accommodation in the area.

Related short courses

Contacts

For further information please email short-course@bristol.ac.uk.

Patches for WinBUGS

From time to time patches are issued, containing fixes for errors in the code, new distributions, new examples, updated manuals and the like. Currently available patches are available below, with full details of what they contain. To apply a patch, open it within WinBUGS and follow the instructions – which follow those for installing the key for unrestricted use.

Winbugs User Manual Version 1.4.3 Free

If you think you have found a bug in the program, please mail details, with examples, to the bugs discussion list – but note that most crashes are caused by users’ coding and modeling mistakes! We also welcome suggestions for other improvements, where these are likely to be of general and practical use.

Patch for WinBUGS 1.4, upgrade to version 1.4.3, dated 6th August 2007

Get the file by clicking this link, and follow the instructions in the file itself.

This patch makes the following changes;

fixed Pareto updating in the case of a conjugate uniform likelihood — if the upper bound of a uniform distribution (on (0, theta), say) is assigned a Pareto prior, then the full conditional for theta is Pareto, but the full conditional parameters were previously calculated incorrectly.
fixed calculation of deviance and prior density (non-conjugate) for “dchisqr” distribution — previously if “dchisqr” had been used as a non-conjugate prior, then its contribution to the full conditional was calculated incorrectly; the contribution to the deviance has always been calculated incorrectly, regardless of conjugacy.
improved Brooks-Gelman-Rubin convergence diagnostics — see “The Inference Menu” –> “Samples…” in the updated manual for details.
updated manual entry on “bgr diag” — see “The Inference Menu” –> “Samples…”
included “Lunn et