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.

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 al” (Statistics and Computing, 2000) reference in “References” section of manual.
• corrected typos in “Rats” example.
• fixed greying-out of “DIC…” on “Inference” menu — previously it was not possible to re-access the “DIC Tool” (if closed) after the DIC monitor had been set.
• + all the changes below contained in Patch 1.4.2.

Patch for WinBUGS 1.4, upgrade to version 1.4.2, dated 13th March 2007

This patch makes the following changes;

• fixed bounding of interval-censored binomial nodes: the lower bound was incorrectly set equal to the upper bound such that the only permissible value was the upper bound itself.
• slightly improved display of “model fit” in “Comparison Tool”
• Fixed a problem whereby if the precision of a Normal distribution was assigned a uniform prior then the parameters of the resulting conjugate, gamma full conditional were calculated incorrectly. This was also an issue if precision = a * b and either a or b were given a uniform prior.
• fixed “spatial.exp”: the correlation function was incorrectly calculated as exp[-phi * (d^kappa)]. It has now been corrected to calculate exp[-(phi * d)^kappa]
• All the changes below contained in Patch 1.4.1.

Patch for WinBUGS 1.4, upgrade to version 1.4.1, issued 22nd Sept 2004

This patch makes the following changes;

• new “About” box to show version number (1.4.1)
• new front page for manual, to show new version number and to provide easy access to this file
• fixed reverse polish problem: e.g. a + (b – c) –> a + c – b (now gives a + b – c)
• new GeoBUGS manual — included version is now called GeoBUGS Version 1.2
• spatial.unipred distribution now works even if the “spatial.exp” or
• spatial.disc model used to estimate the covariance matrix is based on vectors of length > 99
• new “spatial.disc” distribution added
• new “dpois.conv” distribution added
• new multivariate CAR distributions added: “mv.car” and mv.car.uncon
• fixed problem whereby squares couldn’t be selected in “grid maps”
• new GeoBUGS map: “Forest”
• new GeoBUGS map: “WestYorkshire”
• new GeoBUGS map: “Huddersfield_750m_grid”
• fixed sum-to-zero constraint in “mv.car”
• fixed sum-to-zero constraint in “car.normal”
• fixed sum-to-zero constraint in “car.l1”
• fixed problem with identity links (GraphStack and GraphMixture)
• mixture models (GraphMixture) may now have three non-constant indices as opposed to two
• fixed problem with generalised gamma distribution (notified by Nicky Welton)
• fixed “Variable2” syntax error (notified by Andrea Manca)
• fixed problem with “ranked” function (notified by Erik Jørgensen)
• new “inprod2” function added — can be used as an alternative to “inprod” for slow models (run-speed of “inprod2” is comparable to “inprod” in WinBUGS 1.3) (notified by Zhao & Bonizzoni)
• new version of “Blockers” example (with non-gamma priors on precisions)
• new version of “Dyes” example (with non-gamma priors on precisions)
• new version of “Equiv” example (with non-gamma priors on precisions)
• new version of “Inhalers” example (with non-gamma priors on precisions)
• new version of “Kidney” example (with non-gamma priors on precisions)
• new version of “Leukfr” example (with non-gamma priors on precisions)
• new version of “Oxford” example (with non-gamma priors on precisions)
• new version of “Rats” example (with non-gamma priors on precisions)
• new version of “Salm” example (with non-gamma priors on precisions)
• new version of “Seeds” example (with non-gamma priors on precisions)
• new version of “Surgical” example (with non-gamma priors on precisions)

The BUGS Site

• Welcome Page
• Latest News
• Contact us/BUGS list
• WinBUGS
• New WinBUGS examples
• The BUGS Book
• FAQs
• DIC
• GeoBUGS
• PKBUGS
• Running from other software
• BUGS resources online
• WinBUGS development site
• OpenBUGS site