Re: bivariate model with some complications

From: awils15 <asremlforum_at_VSNI.CO.UK>
Date: Tue, 4 Aug 2009 12:31:45 +0100

Many thanks for the suggestion and the explanation. I've tried running your suggested model and I'm hoping you won't mind that I have some follow questions!

Firstly, the Animal Error for trait 1 as modelled by at(Trait,1).ide(FO) is indeed going negative (though it is positive with the covariance term added back in.

 Source Model terms Gamma Component Comp/SE % C
 ide(FO) 4051 4051 0.177122 0.177122 4.64 0 U
 at(Trait,1).ide(FO) 4051 4051 -0.160962 -0.160962 -4.28 0 U
 Residual UnStructured 1 1 0.150491 0.150491 69.65 0 P
 Residual UnStructured 2 1 0.00000 0.00000 0.00 0 F
 Residual UnStructured 2 2 8.92042 8.92042 8.92 0 P
 Trait.FO UnStructured 1 1 0.469560E-02 0.469560E-02 2.35 0 P
 Trait.FO UnStructured 2 1 -0.302223E-01 -0.302223E-01 -0.98 0 P
 Trait.FO UnStructured 2 2 2.26955 2.26955 2.30 0 P

Should I try rescaling perhaps and/or might it be appopriate to fix this second component to be positive (!GP) in this case?

Constraining this component to be positive yields -

Source Model terms Gamma Component Comp/SE % C
 ide(FO) 4051 4051 0.162698E-01 0.162698E-01 6.99 0 P
 at(Trait,1).ide(FO) 4051 4051 0.704107E-07 0.704107E-07 0.00 0 B
 Residual UnStructured 1 1 0.150502 0.150502 69.63 0 P
 Residual UnStructured 2 1 0.00000 0.00000 0.00 0 F
 Residual UnStructured 2 2 8.46321 8.46321 8.34 0 P
 Trait.FO UnStructured 1 1 0.452671E-02 0.452671E-02 2.27 0 P
 Trait.FO UnStructured 2 1 0.405022E-01 0.405022E-01 1.53 0 P
 Trait.FO UnStructured 2 2 2.75059 2.75059 2.57 0 P

So, if I am interpreting correctly the animal error variance estimate for trait 1 is atually similar under the two models (i.e. sum of first two components =approx 0.016). It also - reassuringly - is in agreement with an estimate from univariate analysis. However, the error covariance is obviously an order of magnitude less under the constrained model and I'm not sure how to interpret this.

#################################
Secondly, with regard to the analysis on the binomial scale I am still having difficulty getting ASReml to accept the job. Specifically I am getting the following message reported in .asr.

 The Multivariate Structure assumes I x US R structure
  Set the !ASUV qualifier to use any other R structure.
  You probably also need mv in your model and !S2==1 in your R structures
 Fault 1 REQUIRE !ASUV qualifier for this R structure

Sorry - I suspect that there is a very simple solution to this which I am just missing but any suggestions would be welcome!

-------------------- m2f --------------------

Sent using Mail2Forum (http://www.mail2forum.com).

Read this topic online here:
http://www.vsni.co.uk/forum/viewtopic.php?p=708#708

-------------------- m2f --------------------
Received on Wed Aug 04 2009 - 12:31:45 EST

This webpage is part of the ASReml-l discussion list archives 2004-2010. More information on ASReml can be found at the VSN website. This discussion list is now deprecated - please use the VSN forum for discussion on ASReml. (These online archives were generated using the hypermail package.)