[R-sig-ME] How to assess significance of variance components (Please discard previous e mail... but read this one)

[Stephane Chantepie]

Hi all,

First, thank a lot Greg for the ideas you gave but I am not sure to 
understand really what mean a 'point mass' and to be honest I am really far 
from being a developer.
I think that the alternative approach which consist in defining a threshold 
and considering that all value below this threshold could be considered as 0 
appear to be really suggestive.
Even if I agree with you that testing everything is not always necessary,in 
some cases, it could be important.

For my case : 

I have done random regression animal models to test an increase in genetic 
variance across age. I have tested a VA by AGE interaction (which appear to 
fit the VA estimates I had with univariate animal models computed by 
age).Then, I have tested a quadratic interaction which also does not 'look' so 
bad compare to estimates from univariate models. So, I have then tested a 
cubic interaction (error interval appear pretty big) ... 
For all theses models,parameter posteriors are well shaped, but I am unable to 
know (or decide) if a model VAxAGE fit well my data, and which is the best 
models.
So, in the cases where random part of models is not just a correction, it 
could be interesting to know the significance of a random parameter in order 
to take a decision (and to justify it in a manuscript).

His someone had already been in front of this difficult choice?
How did you deal with this problem?

Thank a lot 
Stephane