[vsni.co.uk] Contact us
Author Message
Post new topic

  ASReml  ~  Mean of breeding values different from 0 / covariance issue

tommy92
Posted: Wed Apr 08, 2009 12:04 pm Reply with quote
Joined: 16 Mar 2009 Posts: 6
Hello everyone,

I have two issues with pedigree analysis.
The first one is that the mean of the Breeding Values for different traits is not equal to zero (even far from 0). Is there a problem?

The command line is like this:
Weight ~ mu Age Sex !r ID ide(ID) YearBirth Year

The second issue is about covariance analysis. When launching the model, it gives an error instead of the matrix of variance/covariance:
"REQUIRE !ASUV qualifier for this R structure"
However, I'm using an "individual" model, whitout any group or family.
Is it a problem with the command line?
Trait1 Trait2 ~ Trait Trait.Age !r Trait.ID Trait.ide(ID) Trait.YearBirth Trait.Year

1 2 4
Trait 0 US
3*0

Trait.ID 2
Trait 0 US
3*0

Trait.ide(ID) 2
Trait 0 US
3*0

Trait.YearBirth 2
Trait 0 US
3*0

Trait.Year 2
Trait 0 US
3*0

Thank you in advance!
View user's profile Send private message
Arthur
Posted: Thu Apr 09, 2009 4:47 am Reply with quote
Joined: 05 Aug 2008 Posts: 471 Location: Orange, NSW
Dear Tommy,

1) BLUPs do not in general sum to zero if there is a correlation structure as in a genetic analysis.

2) The coding is missing a line and should read.


1 2 4
0 # Added
Trait 0 US
3*0

Trait.ID 2
Trait 0 US
3*0
ID 0 AINV # Added

Trait.ide(ID) 2
Trait 0 US
3*0
ide(ID) # Added

Trait.YearBirth 2
Trait 0 US
3*0
YearBirth # Added

Trait.Year 2
Trait 0 US
3*0
Year # Added

However, these last two may not be able to be fitted unless there are
quite a few Years and YearBirth levels. I would normally
fit Trait.Year as a fixed effect (although you may have valid reasons for not doing so).

_________________
Arthur Gilmour

Retired Principal Research Scientist (Biometrics)
View user's profile Send private message Send e-mail Visit poster's website
tommy92
Posted: Fri Apr 10, 2009 10:34 am Reply with quote
Joined: 16 Mar 2009 Posts: 6
Thank you Arthur,

It works well now.
Regarding covariances, i have respectively 10 and 20 levels for year and year of birth. Should I put at least one effect as fixed?

Concerning the correlation structure you mentioned, which could explain a mean of BV different from zero, i'm not sure to understand its origin?

Thank you
View user's profile Send private message
Arthur
Posted: Tue Apr 14, 2009 10:34 pm Reply with quote
Joined: 05 Aug 2008 Posts: 471 Location: Orange, NSW
Dear TOmmy,

"Regarding covariances, i have respectively 10 and 20 levels for year and year of birth. Should I put at least one effect as fixed?" I can't tell you what you should do. Depending on the structure of the data, differences caused be the 'environment' could affect the reported breeding values if say Year effects are fitted as random. Treating Year as fixed ensures there is no 'recovery of "year" information'. It is common for environmental effects to be fitted as fixed, but they are sometimes fitted as random, especially if there are many levels or they are derived on the basis of randomization arguments.

Concerning the correlation structure you mentioned, which could explain a mean of BV different from zero, i'm not sure to understand its origin?

Consider 3 observations: 2 4 6.
Under an independent model, the mean is 4 and the residuals are -2, 0, 2

However, if we add some correlation to the model by saying observations 1 and 2 represent one family, and the 3rd observation represents a second family, we might say that the mean is 4.5, family effects are -1.5 and +1.5 and residuals are -1, +1, 1.5

Now if we make family random, family effects take on values -f, f where f is between 0 (independent model) and 1.5 (fixed model).
For a variance ratio 1.5, mu is 4.333, f is 1.0 so the simple residuals are
-1.333, +.667 +.667

However, if we used a variance structure at the residual level (i.e.
instead of fitting family, we use a correlation matrix to fit the equivalent model),
the residuals would be -2.333, -.333 and 1.667 which do not sum to 0.0

When you use a genetic relationship matrix, you are incorporating the family and individual effects and so the BLUPs do not in general sum to zero.

I trust that helps.

_________________
Arthur Gilmour

Retired Principal Research Scientist (Biometrics)
View user's profile Send private message Send e-mail Visit poster's website
tommy92
Posted: Fri Apr 17, 2009 6:39 am Reply with quote
Joined: 16 Mar 2009 Posts: 6
Thank you very much Arthur for these explanations!
View user's profile Send private message

Display posts from previous:  

All times are GMT
Page 1 of 1
Post new topic

Jump to:  

You cannot post new topics in this forum
You cannot reply to topics in this forum
You cannot edit your posts in this forum
You cannot delete your posts in this forum
You cannot vote in polls in this forum
You can attach files in this forum
You can download files in this forum