Introduction to BRugs, software for calling OpenBUGS from R

• Instructions for older versions of R, which you have previously installed from cran.r-project.org: Simply start R, and download the BRugs package by typing: install.packages("BRugs")

• Instructions for those who are just getting started in R: Follow these instructions, which require you to download R Version 2.11 (*not* the latest version)

• You may want to download the R help files for BRugs, BRugs.pdf

• Download the files for Example 1: Hierarchical beta-binomial model for metaanalysis of success proportions. The data are the results of 9 studies concerning the effectiveness (yes or no) of a medical device; the studies have between 6 and 46 patients. Our goal is to obtain better estimates of the success probability in each study by borrowing strength across studies, and also estimate the predictive probability of success for a patient in a new, 10th study that has not occurred yet.

  WARNING: DON'T save these files by right-clicking in Explorer; instead, open the file and do "File-Save As", *OR* (better yet) open the file and copy-paste it into a .txt file to ensure proper decoding of carriage returns!!!

  • The BRugs code itself, betabinHM_BRugs.txt, OR a newer, more compact version, betabinHMnew_BRugs.txt, that uses only the "bugsData", "bugsInits", and "BRugsFit" commands, AND
  • The WinBUGS code that these R programs call, betabinHM_BUGS.txt
  The data and inits files needed by BUGS are simple enough in this example that they are created by the BRugs code.

  From within R, click "File" and pull down to "Change dir" to change your the working directory to the one where you stored the two above files (you'll see I stored my files in K:\book3). You will also need to edit the "setwd" command at the top of the BRugs file, or simply comment this line out (put a "#" at the beginning of the line).

  Speaking of comments, to save copies of the plots that are created you may want to un-comment out the "postcript" and "dev.off" commands. Indeed, the first time you run this you may want to simply cut and paste the commands into R one line at a time, so you can see what each one does.

• Download the files for Example 2: Computing exact leave-one-out standardized residuals and CPO values for the stack loss data.
  (In this version, I saved the files in K:\book\CL3\BRugs).

  • The BRugs code itself, stacks_BRugs.txt
  • The WinBUGS code that it calls, model.txt
  • The covariate (X) data, dataX.txt; the response (Y) data are in the BRugs file above.
  • The two initial values files, Inits1.txt and Inits2.txt

  From within R, source the stacks_BRugs.txt file. This should call OpenBUGS 21 times (once for each data point), and ultimately obtain a plot of the two diagnostics.

  This example is based on work by Dr. Haijun Ma, and follows the residual analysis portion (pp.33-38) of my intermediate Bayesian data analysis short course slides, which in turn follow Section 2.5 (pp. 79-86) and Exercise 22 (p. 104) of the Carlin and Louis book, Bayesian Methods for Data Analysis , published by Chapman and Hall/CRC Press/Taylor and Francis.

• Download the files for Example 3: Estimating power for a hypothetical clinical trial based on a Weibull survival model.
  (In this version, I saved the files in C:\BRugs\PowerCalcs)

  • The BRugs code, Power_BRugs.txt
  • Three different WinBUGS models that it calls, refmodel.txt, optimistmodel.txt, and skepticmodel.txt

  From within R, source the Power.BRugs file. This file generates the data (which is all fake here, since this is a design problem) and the initial values as well.

  This example follows the clinical trial design portion (pp.39-44) of my intermediate Bayesian data analysis short course slides, which are in turn based on a Bayesian class project paper written by Joe Koopmeiners. The "classic" reference in this area is:

  • Spiegelhalter, D.J., Freedman, L.S., and Parmar, M.K.B. (1994). Bayesian approaches to randomised trials (with discussion). J. Roy. Statist. Soc., Ser. A, 157, 357--416.

• Download the files for Example 4: Estimating power for a drug clinical trial based on a binomial response (alive/dead) model.
  

  This material is all located on a separate website created by Brian Hobbs, current PhD student in the Division of Biostatistics at the University of Minnesota. The work is based on Section 3 of a paper by Hobbs and Carlin, submitted to the Journal of Biopharmaceutical Statistics.

• Download the files for Example 5: Estimating power for a device clinical trial based on a Cox survival model.
  

  This material is also located on Brian Hobbs' website, and described in Section 4 of the Hobbs and Carlin paper.