ASReml-R and using weights with the ped() function

My question is about using a weights argument when the ped() function
is part of the random section of the model.

I'm using pedigree information for progeny testing of various parents
(kiwifruit, not animals). The parents are represented in the progeny
data with these frequencies, not outrageously unbalanced, I'd have
thought:

Code:
> with(model.df, table(Pa, Ma)) -1 # (family BLUPs are to be predicted)

(In the absence of a better way to distinguish output from other text,
I've indicated with the '-:' markers)

Code:
-: Ma
-: Pa S T U
-: 01 24 0 0
-: 02 10 23 18
-: 03 26 25 19
-: 04 15 10 19
-: 05 15 11 11
-: 06 19 17 16
-: 07 17 18 22
-: 08 32 0 0

Without weights, this works fine:

Code:
> discard.asr <- asreml(fixed = Discard_Total ~ 1 + Season,
                      random = ~ped(Vine, var = T) + Block,
                      data = model.df, na.method.X = "include",
                      ginverse = list(Vine = G.inv$ginv))

Code:

-: Equations: 513 (5 dense)
-: Initial update shrinkage factor: 0.316
-: 1 singularities detected in design matrix
-:
-: LogLik S2 DF
-:
-: -1117.0549 152.2097 363 11:27:48 (1 component(s) constrained)
-: -1115.3695 150.3475 363 11:27:48
-: -1114.4721 142.8087 363 11:27:48
-: -1113.6746 129.4189 363 11:27:48
-: -1113.3548 116.7492 363 11:27:48
-: -1113.2546 107.1695 363 11:27:48
-: -1113.2326 101.6364 363 11:27:48
-: -1113.2294 99.2483 363 11:27:48
-: -1113.2290 98.4350 363 11:27:48
-:
-: Finished on: Wed Dec 3 11:27:48 2008
-:
-: LogLikelihood Converged

For every other response analysed, there is a similar message about
the 'singularities', but it's evidently not fatal.

If I add a weights argument, though, I get this:

Code:

-: Equations: 513 (5 dense)
-: Initial update shrinkage factor: 0.316
-: 1 singularities detected in design matrix
-:
-: LogLik S2 DF
-:
-: -1155.4065 4872.1492 363 11:27:49 (1 component(s) constrained)
-: Warning: 1 singularities inAverage Information Matrix
-: Exit status: -1 - Singularity in Average Information Matrix
-:
-: Finished on: Wed Dec 3 11:27:49 2008
-:
-: Error: Abnormal termination
-: Singularity in Average Information Matrix
 

It 'works' with a weights parameter and no pedigree -- again with the
Singularity message.

In searching for reasons why there's such a difference, I set all the
weights to being the same, I'd have thought I'd get the same as the
first one, but no:

Code:
-: Equations: 513 (5 dense)
-: Initial update shrinkage factor: 0.316
-: 1 singularities detected in design matrix
-:
-: LogLik S2 DF
-:
-: -1128.3707 1537.1636 363 11:28:04 (1 component(s) constrained)
-: Warning: 1 singularities inAverage Information Matrix
-: Exit status: -1 - Singularity in Average Information Matrix
-:
-: Finished on: Wed Dec 3 11:28:04 2008
-:
-: Error: Abnormal termination
-: Singularity in Average Information Matrix

I'm not surprised by the fact that the initial S2 is about 10 times
the corresponding one when weights weren't specified since the weights
were all set to 10. My question is: why should the singularity be a
problem when there's putative weighting, but not when it's not
mentioned?

I've tried setting initial values, but since nothing made a
difference, I thought it prudent to check the more basic story first.
It seems as though I'm not understanding something vital. If it
won't work for equal weights, it won't work for the real ones. I've
also tried omitting Block and Season from the model, but to no avail.

If it's not just something silly I'm doing with the syntax, I can
supply a .RData type file with the model.df from which the rest is
calculated so others can replicate what I've tried.

Other information that might be of consequence:

Code:
> library(help = "asreml")

-: Information on package 'asreml'
-:
-: Description:
-:
-: Package: asreml
-: Title: asreml() fits the linear mixed model
-: Version: 2.00
-: Author: David Butler
-: Description: asreml estimates variance components under a general
-: linear mixed model by residual maximum likelihood.
-: Maintainer: David Butler <david.butler_at_dpi.qld.gov.au>
-: License: VSN International
-: Packaged: Tue Aug 28 12:28:20 2007; butlerd
-: Built: R 2.5.0; ; 2007-08-28 12:28:21; unix
-:
-: [..etc.,.]
-:

> sessionInfo()
-: R version 2.8.0 (2008-10-20)
-: i686-pc-linux-gnu
-:
-: locale:
-: LC_CTYPE=en_US.UTF-8;LC_NUMERIC=C;LC_TIME=en_US.UTF-8;LC_COLLATE=en_US.UTF-8;LC_MONETARY=C;LC_MESSAGES=en_US.UTF-8;LC_PAPER=en_US.UTF-8;LC_NAME=C;LC_ADDRESS=C;LC_TELEPHONE=C;LC_MEASUREMENT=en_US.UTF-8;LC_IDENTIFICATION=C
-:
-: attached base packages:
-: [1] grid grDevices utils stats graphics methods base
-:
-: other attached packages:
-: [1] asreml_2.00 lattice_0.17-15

(A lot has happened in the R world since the asreml package was
compiled. The oldest version I still have is R-2.6.2 which gives the
same results.)

Thanks

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

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

Read this topic online here:
http://www.vsni.co.uk/forum/viewtopic.php?p=324#324

-------------------- m2f --------------------
Received on Fri Dec 03 2008 - 01:37:39 EST