[R-sig-ME] MCMCglmm

Jarrod Hadfield j.hadfield at ed.ac.uk
Sun Feb 1 13:58:03 CET 2009


Thanks to everyone who emailed on or off-list about MCMCglmm. I had a  
lot of suggestions about additions/improvements and I've uploaded an  
update on CRAN with most of the teething problems sorted out. This  
should be available over the next couple of days

For those experiencing  dramatic crashes this is caused by variances  
hitting zero or correlations going to -1/1.  This problem does not  
happen when a proper prior is specified, which is recommended.   
MCMCglmm should terminate with more class in the updated version, and  
should print a suitable error message.

Users also asked about the negative binomial distribution, Bayes  
Factors/DIC, and alternative link functions.

The Poisson log-normal seems to fit many data as well as the neg bin  
so it probably wont be included in the near future.

I'm about half way through the code for obtainging DIC

Alternative link functions are easy to implement and if you email me  
the link function you would like and for which distribution, I'm happy  
to include it.

Alternative single parameter dsitributions are also easy to implement  
and I'm happy to include them if you can send me/refer me to their pdf  
and cdf.  If you already have C/C++ code to calculate the density  
functions then even better.



-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: not available
URL: <https://stat.ethz.ch/pipermail/r-sig-mixed-models/attachments/20090201/1a12a841/attachment.pl>

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