[R-sig-ME] MCMCglmm 2.05 update

Jarrod Hadfield j.hadfield at ed.ac.uk
Tue Aug 10 22:16:12 CEST 2010


I've released a new version of MCMCglmm with the following:


I) Sparse design matrices with certain types of contrast were  
sometimes incorrect (e.g sum contrasts when only a two-level factor  
was fitted in the fixed term)

II)  Latent variables for censored Gaussian data were sometimes  
updated incorrectly


I) with idh structures the prior for each diagonal element was  
equivalent to the marginal distribution of the variance under a  
standard inverse-Wishart distribution. This seemed to confuse many so  
now the prior specification for the ith variance is simply a  
univariate inverse-Wishart with V*=V[i,i] and nu*=nu


I) summary function which includes the dreaded P-value (although I  
wouldn't really call it that)

II) hurdle Poisson ("hupoisson") zero-truncated Poisson ("ztpoisson")  
and zero-altered Poisson ("zapoisson" )distributions can now be fitted  
- these models have much better mixing properties than  zero-inflated  
Poisson models ("zipoisson") and they are discussed a little in the  

III) predict function with confidence and prediction intervals on the  
data and link scale.  This is not yet implemented for all  
distributions and should be treated with care.

IV) response-feedback models  - developmental only

Thanks to everyone who sent bug reports and offered advice,  it is  
very much appreciated.



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