Dear BUGS users,
Couple of days ago I posted a query regarding an error message when I
attempted to use multivariate t-distribution. I would like to thank David
Spiegelhalter and Mike Thompson for their reply and I found the answers
very helpful.
Here are the replies:-
1. Section 3.4 of the WinBUGs manual states that Wishart priors can only
be used as conjugate priors for multivariate normals. So you cannot
implement the multivariate t in this way.
You might implement it as a sequence of univariate conditional
distributions.
We will try to put multivariate t in the next version.
David
2. The key to solving your problem is to use Sigma.alpha directly in the
dmnorm()
distribution and scale your alpha values afterwards.
I've revised your file to reflect this, ran it and included the results
with the
program in the attached file.
CAUTION: Make sure that I've got the scaling and semantics correct.
(See attached file: imr-poissin-t-revised.odc)
-Mike Thompson
P&G, Specialist in Mathematical Modeling and Statistics
*****************************************************
* Md. Monir Hossain, Ph.D. *
* Research Fellow *
* Department of Public Health and General Practice *
* Christchurch School of Medicine & Health Sciences *
* University of Otago *
* PO Box 4345 *
* Christchurch, New Zealand *
* Phone: 64-3-364-0450 Ext. 4039 (Off.) *
* 64-3-348-2001 (Res.) *
* Fax: 64-3-364-0425 (Off.) *
* E-mail: [log in to unmask] *
*****************************************************
|