Re: error structure in bivariate models with ASReml3

Dear Anne,
1) '3130 records where both LD, AD and SEX are known'

But the data summary shows 1373 AD values are missing!

2) AD LD ~ Trait Trait.SEX with !MVINCLUDE
   will use all data (AD LD values) properly accounting for
   missing values in these response traits, and treating
   any missing SEX values as an extra SEX class (0).

   !MVINCLUDE is the default in ASReml 1. In ASReml 2 and 3,
   it must be specified explicitly to achieve this result.

3) I note LD is declared as a factor. This might result in different
results in ASReml 3 cf ASReml 1.

4) Were there 612 records with missing Sex. If not, I would fit the
model using all records with known sex.

5) The incomplete output from ASReml 3 looks like a program crash.
If you send me (arthur.gilmour_at_cargovale.com.au) the data (and .as file)
I can work out why, and fix the problem.

6) "“6260 records assumed pre-sorted 2 within 3130” : this means
that the two traits in the bivariate model have both 3130 values
recorded, is that right?"
   NOT QUITE. Yes, there are 3130 records for each trait, but some
could be 'missing values'. It is saying that ASReml has taken two
responses (AD and LD) from each of 3130 data records and they are
arranged as 3130 pairs so that the error variance matrix is defined as
a direct product of [Identity of order 3130] cross [Unstructured matrix
of order 2]

7) "- it seems I no longer need the !S2==1 option which I still use in
my univariate models, else I get an extra error variance. Why do I not
need it now?"

For a single set of univariate data, ASReml by default factors out the
variance, so the default R structure is a correlation matrix and the
variance components for the random terms are estimated as GAMMA ratios.

However, you can define aaan R structure which includes a variance
parameter. Then you need !S2==1 to make the scaling parameter 1.
[Then The GAMMA column and the VARIANCE column in the output will
agree.]

For multivariate data, and multiple sets of data, since each trait and
set is expected to have a different variance, all parameters are
estimated on the Variance scale.

This can be confusing syntax wise. We have attempted to improve the
description of !S2== in the latest edition of the user guide.

>

On Tue, 2009-12-22 at 16:26 +0100, Anne CHARMANTIER wrote:
> Thank you for your swift answer. The fact is, I have really 3742
> records, and I have not used !ASUV, but I guess ASReml was counting
> 7364 records when amounting all records where LD was available, and
> all where AD was available. Your option unfortunately gave the same
> error message. However, since you pointed out that this was probably
> because of missing records, I restricted the dataset to the 3130
> records where both LD, AD and SEX are known, and the model runs
> correctly, and converges. Below is the full output from the .asd.
> Just a couple of extra questions if you have the time, to make sure I
> am in the right track:
> - “6260 records assumed pre-sorted 2 within 3130” : this means that
> the two traits in the bivariate model have both 3130 values recorded,
> is that right?
> - it seems I no longer need the !S2==1 option which I still use in my
> univariate models, else I get an extra error variance. Why do I not
> need it now?
>

-- 
Best Wishes,
Arthur Gilmour
Adjunct Professor
School of Computing and Mathematics
Charles Sturt University
Joseph inherited  the title  'King of the Jews',  being the leading 
descendent of King David. However, neither the Romans or King Herod 
were about to  recognize him as such.  After he contracted to marry 
Mary,  he discovered  she  was  pregnant.   He  was  surprised  and 
disappointed,  and contemplated a quiet divorce;  to proceed  would 
confer his  title  on her son.  But an angel  appeared  to him in a 
dream saying,  "Joseph, Son of David,  do not be afraid  to take to 
you Mary,  your wife,  for that which is conceived in her is of the 
Holy Spirit.  She will have a son whom you shall name JESUS, for He 
shall save His people from their sins."
When Jesus was born, Herod killed all the babies of Bethlehem under 
two years but  Joseph saved Jesus,  taking Him to Egypt until Herod 
died.
Ultimately,  the Romans  crucified  Him with  the charge  'Jesus of 
Nazareth,  the King  of the Jews'.  But He rose  from the  dead the 
following Saturday night  and is now waiting in Heaven for the time 
God has set for Him to  return  and take up His title and throne in 
Jerusalem.  
 
Mobile Number +61 427 227 468
Home phone +61 2 6364 3288  Skype: Arthur.Gilmour
http://www.CargoVale.com.au/asreml
Travel:
Touring S E Asia 29Dec - 6 Feb
WCGALP 2010   Aug 1-6