Covariate analyses

# Covariate analyses

```A PhD student is analysing some sheep data.  The traits in question are two
scrotal a circumferences.  She would like to fit age (coded as dob) and live
weight as covariates.  The two options below work and give 'reasonable'
results.  The first option is as shown in an example in the ASREML Manual.
The second option uses lin( ).  The two options give very similar results, but
they are not identical.  Why is this so?  Is there a prefered option for this sort of
situation?

First option

Animal Model
animal     !P
sire
dam
year     4 !I
stud     4
aod      4 !I
tob_r    5 !I
dob      1
lw       1
sc5
sc10
sc16

c:\asr\ped5.dat
c:\asr\ped5.dat

sc5 sc10 ~ Trait Tr.year Tr.stud Tr.aod Tr.tob_r Tr.dob Tr.lw !r Tr.animal
1 2 1                   # 1 set of resid, 2 dim anim.trait, G str tr.animal
0                       # use all records
2 0 US 8.11 3.98 8.79   # no. of traits, records sorted,US, residual ervar1 erco12
ervar2

Tr.animal 2             # 2 dim struct, always 2
2 0 US 0.90 0.27 0.69   # 2 traits, sorted, US, animal var1 cov12 var2
0                       # use all records

Second option

Animal Model
animal     !P
sire
dam
year     4 !I
stud     4
aod      4 !I
tob_r    5 !I
dob      1
lw       1
sc5
sc10
sc16

c:\asr\ped5.dat
c:\asr\ped5.dat

sc5 sc10 ~ Trait Tr.year Tr.stud Tr.aod Tr.tob_r lin(dob) lin(lw) !r Tr.animal
1 2 1                   # 1 set of resid, 2 dim anim.trait, G str tr.animal
0                       # use all records
2 0 US 8.11 3.98 8.79   # no. of traits, records sorted,US, residual ervar1 erco12
ervar2

Tr.animal 2             # 2 dim struct, always 2
2 0 US 0.90 0.27 0.69   # 2 traits, sorted, US, animal var1 cov12 var2
0                       # use all records

Dr Raul W. Ponzoni
Principal Research Scientist (Livestock Genetics)
South Australian Research and Development Institute
G P O Box 397