For a scalar beta,
beta ~ dnorm(zero, tau) I(betalower, betaupper);
will produce gibbs sampling variates that only fall within betalower
and betaupper. I need to do a multivariate version of I() as explained
below.
In a data analysis problem involving correlated censored variables, I
need to produce VECTOR gibbs sampling variates that are constrained to
fall within hyperrectangles. I would like to do something like:
beta[] ~ dmnorm(zero[], omega[,]) I(betalower[], betaupper[])
where beta[] is now a vector, betalower[] and betaupper[] now mark the
limits of the acceptable hyperrectangle. The above statement does not
work as is, presumably because I() does not handle vectors.
Is there any way in BUGS to specify acceptable regions on VECTOR
nodes?
Thank you,
- Anand
-----------------------------------------------------------------
Anand V. Bodapati Rm 492, Leverone
Assistant. Professor Phone: (847)467-2767
Kellogg Graduate School of Management Sec'y: (847)491-3522
Northwestern University Fax: (847)491-2498
Evanston, IL 60208-2001 Email: [log in to unmask]
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|