Problem with LogL when fitting genetic group

Hello Arthur,
I have followed your instructions by adding an extra genetic group for each
breed of pigs in the pedigree file.
LogL values are still not comparable. The problem will go away when genetic
group is not fitted (LogL values make sense and they are comparable).

In my data analysis, I want to test whether my genetic correlation estimate
is significantly different from unity (1.00 or 0.99), by comparing LogL
values from the two models; one fixing the rg at 0.99 and the other is to
estimate the rg. The results I have are as follows;

WHEN FITTING "GENETIC GROUP"
1. Let the program estimate rg (which we expect LogL to be maximum)
Tr 0 CORR 0.99 2402.52 1662.95 !GP
Results: LogL = -44702.7; rg = 0.7582; Var1 = 2409.29: Var2 = 1654.41
2. Fix rg = 0.99
Tr 0 CORR 0.99 2402.52 1662.95 !GFPP
Results: LogL = -44662.9; rg = 0.99; Var1 = 2400.43; Var2 = 1460.81
(This LogL does not make sense because it is higher than the one from the
model letting rg go free)

WITHOUT "GENETIC GROUP"
1. Let the program estimate rg
Tr 0 CORR 0.99 2402.52 1662.95 !GP
Results; LogL = -44721.7;rg = 0.6986; Var1 = 2862.37;Var2 = 1432.21
2. Fix rg = 0.99
Tr 0 CORR 0.99 2402.52 1662.95 !GFPP
Results; LogL = -44721.9; rg = 0.99;Var1 = 2859.11;Var2 = 1060.24
(LogL is as expected because it is lower than the one when rg is not fixed)

You may come up with a new solution to add to your new version of ASReml to
make LogL comparable when fitting genetic group.

Regards

Sansak Nakavisut
PhD student
AGBU, UNE, Armidale, NSW 2351
AUSTRALIA
Tel. 02 6773 3791 Fax. 02 6773 3266
Received on Wed Feb 24 2005 - 13:56:13 EST