Hello,
Your help would be appreciated on the following two questions:
Question 1.
-----------
Could anyone tell me why PROC MIXED _sometimes_ issues positive values of
"log-likelihood'?
I am using the following model:
CLASS locus1 locus2 family;
MODEL trait=locus1|locus2;
RANDOM family locus1*family locus2*family locus1*locus2*family;
and, depending on the trait being analysed, I get positive or negative
'Log-likelihood' values (with the same set of individuals).
Question II.
------------
In order to test for the variance associated with the locus1*locus2*family
interaction being zero (which I want to test with a Likelihood ratio (LR)
test, see Wolfinger's book on PROC MIXED), I also run the following
restricted model:
CLASS locus locus2 family;
MODEL trait=locus1|locus2;
RANDOM family locus1*family locus2*family;
Sometimes, the likelihood of the full model is lower than that of the
restricted model, in which case the LR-statistic takes a negative value! Is
there an explanation? I would expect the likelihood of the full model to be
equal or higher to that of the restricted model; am I wrong?
Your comments will be appreciated,
Xavier.
__________________________________________
!!! New address on October 1st 1999 !!!
__________________________________________
Xavier DRAYE <[log in to unmask]>
Plant Breeding - Crop Science
AGRO/BAPA/ECOP-GC
University of Louvain
Croix du Sud, 2 / 11
1348 Louvain-la-Neuve
BELGIUM
TEL +32 10 472092 FAX +32 10 472021
___________________________________________
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|