[R-sig-ME] MCMCglmm and contrasts

Jarrod Hadfield j.hadfield at ed.ac.uk
Sat Jan 29 15:56:25 CET 2011


Some versions ago there was a problem if contrasts were specified  
where the sum of the contrasts for a factor was zero. This had been  
fixed in version 2.05. In addition, is it possible that some contrast  
cannot be estimated and are dropped? You should get a warning message  
when this happens. You can estimate the contrast by specifying  
singular.ok= TRUE in the call to MCMCglmm, but be careful  - all the  
information is coming from the prior.

Not sure what is meant by the second question. Do you mean you would  
like to obtain some summary measure or summary p-value for all effects  
of the same type? In short, I have not implemented a method for this,  
but occasionally I use Wald tests to get something that resembles a  
p-value. This is almost certainly not technically valid, but on a  
practical basis it seems to give good answers for the few cases I've  



Quoting Szymek Drobniak <geralttee at gmail.com>:

> Hello everyone again,
> it seems this evening is reach in MCMCglmm-based thinking, sorry for
> overloading the list with too much MCMCstuff. My question is - does MCMCglmm
> recognize "contrasts" attribute of data-frames correctly? I've tried fitting
> models to frames with defined contrasts but - at least sometimes it seems
> it's not working (output does not contain all constrasts I've specified).
> Have anyone tried this? And second question, loosely related to previous one
> - is it any way to transform MCMCglmm output (with coeficient for all-1
> levels of the fixed effect) into simple anova output (with one row for each
> fixed effect)?
> Cheers,
> sz.
> --
> Szymon Drobniak || Population Ecology Group
> *Institute of Environmental Sciences, Jagiellonian University
> ul. Gronostajowa 7, 30-387 Kraków, POLAND
> *tel.: +48 12 664 52 19 fax: +48 12 664 69 12
> www.eko.uj.edu.pl/drobniak
> 	[[alternative HTML version deleted]]

The University of Edinburgh is a charitable body, registered in
Scotland, with registration number SC005336.

More information about the R-sig-mixed-models mailing list