Hi,
I know that users have previously reported WinBUGS crashing when the Poisson
is used to model data containing large counts, but I have encountered a
problem that is more subtle to detect. WinBUGS doesn't crash, the crude
diagnostic plots look OK (didn't try multiple chains as I was in early stage
of modelling the process), but the answers were clearly wrong on close
inspection. I'm guessing that in a more complex model the problem might not
have been so easy to identify.
Essentially I was fitting a model like the Alli eg (using
multinomial-Poisson transformation), but with only one categorical factor
(the problem wasn't apparent when a "null model" was fitted, ie no cat
factor). A few of the counts were over 1000, the majority being in the mid
100s. My workaround was to replace the Poisson distribution with a Normal
approximation. Referring to the Ali eg this would be replacing:
X[i,j,k] ~ dpois(mu[i,j,k])
with
X.tau[i,j,k] <- 1 / mu[i,j,k]
X[i,j,k] ~ dnorm(mu[i,j,k], X.tau[i,j,k])
Apologies if this has already been identified (I must have missed it). If
there is a better way to model such data when the counts are relatively
large (the smallest in my case are about 50), I would be interested.
Perhaps more importantly, are there any plans to modify the algorithm for
the Poisson so that it avoids these apparent numerical problems?
Hope this is of interest to at least one other.
> Phil Woodward
>
>
LEGAL NOTICE
Unless expressly stated otherwise, this message is confidential and
may be privileged. It is intended for the addressee(s) only. Access to
this e-mail by anyone else is unauthorised. If you are not an
addressee, any disclosure or copying of the contents of this e-mail or
any action taken (or not taken) in reliance on it is unauthorised and
may be unlawful. If you are not an addressee, please inform the sender
immediately.
Pfizer Limited is registered in England under No. 526209 with its
registered office at Ramsgate Road, Sandwich, Kent CT13 9NJ
-------------------------------------------------------------------
This list is for discussion of modelling issues and the BUGS software.
For help with crashes and error messages, first mail [log in to unmask]
To mail the BUGS list, mail to [log in to unmask]
Before mailing, please check the archive at www.jiscmail.ac.uk/lists/bugs.html
Please do not mail attachments to the list.
To leave the BUGS list, send LEAVE BUGS to [log in to unmask]
If this fails, mail [log in to unmask], NOT the whole list
