This usually rather dormant list has suddenly had a lot of traffic
from people having trouble with their programs.
If your BUGS or WinBUGS program simply won't work, PLEASE could
you try us first on [log in to unmask] and we shall see what we
can do.
In response to some recent questions:
> I am trying to use a dirichlet distribution as a prior for parameters used
> later in a multinomial distribution.
> The model compiles, but when I try to run the model it crashes out with the
> error Trap 126.
> What does this mean and how can I avoid it?
>
Sorry, Dirichlets can only be used as conjugate priors for
multinomials (see Section 3.4 of manual).
>
> Wondering if there is a version beyond BUGS 1.3 which will allow
> multivariate modeling with missing data?
Sorry again, this is not one of the many wonders planned for 1.4. But we will
keep it in mind.
> I tried to run a simple probit regression on Bugs and then WinBugs. On
> Bugs on Linux it runs fine after a total of 500000 iterations. But on
> Winbugs soon after running it stops due to "linear predictor too large".
> An overflow, I guess. There is probably nothing wrong with my code, since
> changing the link from probit to logistic works fine.
Sorry yet again -the difficulty with probit is discussed in Section 9.3
>
> g1=eta11+eta12*z2+eta13*z3
> g2=eta21+eta22*z2+eta23*z3
>
> g1 ~ N(zeta*eta, sigma1)
> g2 ~ N(zeta*eta, sigma2).
>
> The program I have written fails at the "compile" step. So I would be
> pleased if anyone could direct me to some literature, or some examples I
> could use to correct my BUGS code.
>
This looks like you are trying to give a distribution to a function
of variables. I'm afraid this goes against the whole way in which
BUGS is constructed - see Section 9.1 a
David
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|