[R] R2WinBugs and Compare DIC versus BIC or AIC

[Uwe Ligges]

Dae-Jin Lee wrote:
> Dear All
> 
> 1)
> 
> I'm fitting spatial CAR models
> 
> using R2Winbugs and although everything seems to go reasonably well (or I
> think so)
> 
> the next message appears from WINBUGS 1.4 window:
> 
> gen.inits()
> Command #Bugs: gen.inits cannot be executed (is greyed out)

"greyed out" refers to the corresponding GUI. In this case it could mean 
that either your preparations before were not successful or you have 
already specified all initial values and no further have to be 
generated. But this is a WinBUGS question and not R related.

Perhaps you have transposed some matrix? This is always extremely 
confusing ...


> The question is if this message means that something is wrong and the
> results are consequently wrong, or Can I assume it as a simple warning
> message???...
> 
> I've tried to fit the model using just WinBugs (not calling form R) and
> gen.inits command is available and the results obtain are practically the
> same...
> 
> 2)
> 
> The other question is, once several bayesian CAR models are fitted I use DIC
> to model selection
> 
> How can I compare DIC with AIC or BIC obtain for a spatial glm? I mean for
> example:
> 
> - DIC of 300 and Pd of 32
> 
> versus
> 
> - BIC of 220 and Effective dimension (measured of trace of hat matrix) equal
> to 14
> 
> * I've read that DIC is intended as a generalisation of Akaike's Information
> Criterion (AIC), but is it possible to compare them simply looking which is
> the lowest???


Well they are intended to be compared (DIC with DIC, AIC with AIC, but 
do not mix them!) that way given you are comparing within the same class 
of models. And then, there is a bit of fortunetelling with all these 
information criterion, at least in my personal opinion (and I know that 
others disagree).

Uwe Ligges


> 
> Thanks,
> 
> Dae-Jin Lee
> 
> 	[[alternative HTML version deleted]]
> 
> ______________________________________________
> R-help at stat.math.ethz.ch mailing list
> https://stat.ethz.ch/mailman/listinfo/r-help
> PLEASE do read the posting guide http://www.R-project.org/posting-guide.html
> and provide commented, minimal, self-contained, reproducible code.