On 05.09.2011 16:08, Kehl D?niel wrote:> Dear Community,
>
> I know this is not the place to ask WinBUGS questions, but I did not get
> any answers on other lists.
> I am rather new to the BUGS language and to bayesian modeling, excuse me
> for probably simple questions.
> I have to conduct a bayesian meta-analysis of some data. We have
> collected observational and randomized studies related to a certain
> field of interest.
> The idea is to analyse the randomized studies with two different priors.
> One is non-informative, the other is calculated from the observational
> ones. We also want to use a sceptical prior.
> The code I used for the non-informative prior analysis and to get the
> other prior is following:
>
> model
> {
> for( i in 1 : Num ) {
> rc[i] ~ dbin(pc[i], nc[i])
> rt[i] ~ dbin(pt[i], nt[i])
> log(pc[i]) <- mu[i]
> log(pt[i]) <- mu[i] + delta[i]
> mu[i] ~ dnorm(0,1.0E-5)
> delta[i] ~ dnorm(d, tau)
> }
> d ~ dnorm(0,1.0E-6)
> tau ~ dgamma(0.001,0.001)
> sigma <- 1 / sqrt(tau)
> relr <- exp(d)
> }
>
> which appears to work fine after loading data and initials. (there was a
> study with 0 treated and 0 control cases, I had to exclude that one for
> some reasons, is there a solution for this?)
> If I understand right, I can interpret the "relr" as bayesian
estimate
> of relative risk, with credible interval etc.
> I have some questions in connection with the informative prior analysis:
> - after running this same code for the observational data, how do I
> change the specification of d and tau?
> - how can I get posterior probabilities like relr>1?
> - usually how many iterations, thin etc. do we use?
> - can I get nice graphics with both priors and posteriors on it?
>
> I do have to learn everything on my own, so any help is greatly
> appreciated.
> I know R and the BUGS package are able to communicate, is anybody can
> help to solve the task through the R interface would be great.
For model bulding and verification, I recommend to use BUGS directly.
The interface is nice for running estimation processes and comparing
models, not for building them.
Uwe Ligges
> Thank you for you answer or any kind of help:
> Daniel
>
> ______________________________________________
> R-help at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-help
> PLEASE do read the posting guide
> http://www.R-project.org/posting-guide.html
> and provide commented, minimal, self-contained, reproducible code.