Summary of replies (please see the original email below):
-----------------------------------
Andrew Thomas:
I thought WinBUGS / OpenBUGS was not able to calculate DIC for
mixture models. Are you sure that you can get WinBUGS to calculate the
plugin deviance? The deviance will have the component indicator as a parent.
Using the mean of this discrete variable in the plug-in deviance does not
make sense to me.
-----------------------------------
Polychronis P. Kostoulas:
I can tell you how to obtain DIC in WinBugs. First you specify the parameters you wish to trace. Then you run the chain. As the chain runs you select the DIC tool and press set. When the chains ends you may calculate DIC by pressing DIC. DIC is based on all mixture components no matter what parameters you have decided to trace. I hope this helps.
-----------------------------------
Nichole Carlson:
I cannot answer to number 1, but Matthew Stephens has one or more papers on label-switching. I think that he is at U Washington Seattle and has a web page. Sorry that I do not have the references handy.
I also think that the method proposed for number 2 is acceptable based on my own applications and some other applications that I have read. Stephens presents algorithms to resolve label-switching that are more sophisticated than just say, sorting the components at each step.
----------------------------------
Thank you very much for replying.
The website of Matthew Stephens is: http://www.stat.washington.edu/stephens/
Verena
________________________________
Von: (The BUGS software mailing list) im Auftrag von Verena Schmittmann
Gesendet: Mi 18.01.2006 22:02
An: [log in to unmask]
Betreff: [BUGS] comparison of mixture models + label switching problem
Dear list members,
I have some questions concerning the fitting of mixture models with WinBUGS:
1) comparison of mixture models:
On which characteristics of the mixture models does the usefulness of DIC for model comparison in the case of mixture models depend? Which alternatives could I use to compare (well separated) mixtures of time series models? Would the DIC be useful to compare different mixture models which have the SAME number of components? If so, how could I obtain the DIC in WinBUGS for a mixture model?
2) label-switching problem:
Is it true that non-overlap in the posterior distributions of parameters of components of a mixture, and unimodality of the posterior distribution of the mixing proportion may be considered as evidence against the occurence of label-switching?
Thanks very much for sharing references, suggestions, experience, answers..
Verena
-------------------------------------------------------------
Verena Schmittmann
Department of Developmental Psychology <http://www.fmg.uva.nl/op/op_aktueel.cfm>
University of Amsterdam <http://www.uva.nl/>
Roetersstraat 15
1018 WB Amsterdam
The Netherlands
http://users.fmg.uva.nl/vschmittmann/
-------------------------------------------------------------------
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
-------------------------------------------------------------------
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
|