Re: [Fwd: asreml]

From: <arthur.gilmour_at_DPI.NSW.GOV.AU>
Date: Tue, 6 Feb 2007 14:08:32 +1100



I'm fitting a model with an animal effect and a permanent environment
effect. Individuals are spread over two time periods and I want to
test whether a model with time specific animal and PE effects is
better than a model where both periods are treated equally. For
univariate models this is fine:

T1 ~ mu time !r at(time,1).animal at(time,2).animal at(time,
1).identity at(time,2).identity


T1 ~ mu time !r animal identity


1. The univariate models reported assume repeated measures
2. The comparison across time assumes no genetic links between times.
it would be better to include a correlation.
3. Also, if you are allowing different genetic and PE variances, it would
be better
to also allow different residual variances.

Consequently, my version of the first model might end up being

T1 ~ mu time !r time.animal time.identity !uni(time,2) !GU
0 0 2
time.animal 2
time 0 US !GP # assuming genetic links, else DIAG
.1 .01 .1
animal 0 AINV

time.identity 2
time 0 DIAG .1 .1 !GP

You further wrote
If I fit a multivariate model either the animal or the PE effects are
flagged as singular depending on which order the terms are fitted. I
am confident that these parameters can be estimated from the data
because a) all (3) traits have the same pattern of missing data so if
the univariate model is idenitifiable the multivariate model should
be and b) I can fit a multivariate model for each time period
separately by generating two data sets and all model parameters are
estimated. One potential issue is that some individuals are present
in both time periods, but I can't see why this should cause

The most likely cause is that ASReml may not associate the A inverse with
the Tr.time.animal term
when written as a G structure unless it is explicity declared in the
appropriate fashion (see eaxample above).

My attempt at your model is

T1 T2 T3 ~ Trait time Tr.time !r Tr.time.animal Tr.time.identity
2 2 2
zzz # Insert number of records at time 1 after sorting the data
Trait 0 US !GP

zzz # Insert number of records at time 2 after sorting the data
Trait 0 US !GP

Tr.time.animal 3
Trait 0 US !GP
time 0 CORUH !GPPF
.1 1 1
animal 0 AINV

Tr.time.identity 3
Trait 0 US !GP
time 0 CORUH !GPPF
.1 1 1

This is a fairly sophisticated model, so may need better starting values
than I've given.
If you only have a little bit of data, it may be too sophisticated for
your data.
But, according to my understanding of your email, this is what you are
moving towards.

May you know Jesus Christ and His blessing in 2007,

Arthur Gilmour, His servant .

PS My Mixed Model Regression Mapping paper is now available at

Personal website:,
Principal Research Scientist (Biometrics)
NSW Department of Primary Industries
Orange Agricultural Institute, Forest Rd, ORANGE, 2800, AUSTRALIA

fax: 02 6391 3899; 02 6391 3922 Australia +61
telephone work: 02 6391 3815; home: 02 6364 3288; mobile: 0438 251 426

ASREML 2 is now available from
The ASReml discussion group is at
To join it,

Proposed travel:
ASReml workshop (Forestry) 16-18 April Tasmania
Biom Brch meet 2-4 May
ASReml workshop (Animal Breeding) 8-26 May South Africa
AAABG 24-26 September Armidale

This message is intended for the addressee named and may contain
confidential information. If you are not the intended recipient or
received it in error, please delete the message and notify sender. Views
expressed are those of the individual sender and are not necessarily the
views of their organisation.
Received on Sat Feb 06 2007 - 14:08:32 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.)