Skip to content
Prev 2391 / 20628 Next

Mixed model (with interaction) for gene expression and iteration

On Tue, Jun 2, 2009 at 7:53 AM, Paolo Innocenti
<paolo.innocenti at ebc.uu.se> wrote:
Not really.  m2 and m3 are different parameterizations of the same
model but m1 is different.  In models m2 and m3 the random effects for
F and for M are allowed to be correlated within a line.  Thus there
are three parameters determining the variance-covariance of the random
effects.  In model m1 the unconditional distribution of the random
effects for the sex:line interaction is independent of the
distribution of the random effects for line.
I would use either m1 or m3 because I find the correlation structure
in m3 easier to understand than that of m2.
I think so.  I would definitely look into using refit to update the
fitted model rather than starting from scratch with each of the genes.

I would use the idiom
sex:line        line
0.002323722 0.016939242

to get the estimated variances.