Re: LogL with additional random effect
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: LogL with additional random effect

Dear Ron,
The problem is that the updates are trying to move to a point
with a negative component for PE but ASREML keeps forcing PE to be positive.
Thefore, the model is not at the REML solution for PE=0
[unless the PE term has been fixed which is what I try and make it do
but not always successfully.]

SO if PE goes to zero, drop it and refit to get the right solution.


>I have recently run some analyses which were:-
> univariate
> individual animal model
> mean + animal + group effects (group fixed)
>I then added in a permanent environmental effect
>due to dams. For a number of the traits, the ide(dam)
>effect went to zero. However, the LogL was slightly
>lower than for the initial analysis, and the animal
>genetic variance increased. I tried running the analyses
>further with !continue, but nothing changed.
>I presume this is a result of fixing the variance at
>the parameter boundary. Should I be concerned? If I
>saw these results in a paper I would wonder about them.