Hi again,
In addition to my questions below, a treatment was applied in 1997, and a
different experiment was run in 1998. I would assume that if any of the
treatments were actually effective then a significant year*animal effect
may occur given that only some animals were treated. In checking this vague
thought it seems that removing treatment from the model helps convergence
for some traits (ie the model was probably over-parameterised). In light of
this, is there really any sense in checking for a year*animal
interaction....?
Cheers
Kim
>X-Authentication-Warning: lamb.chiswick.anprod.csiro.au: petidomo set
sender to asreml-owner@lamb.chiswick.anprod.csiro.au using -f
>Date: Wed, 02 Feb 2000 15:49:01 +1100
>To: asreml@chiswick.anprod.csiro.au
>From: Kim Bunter <kbunter@metz.une.edu.au>
>Subject: this weeks query is...?
>Sender: asreml-owner@lamb.chiswick.anprod.csiro.au
>
>Hi there,
>
>I am trying to develop appropriate fixed effects models for a variety of
>traits (what's new).
>
>There are relatively few animals involved although up to 8 repeated records
>(8 production years) per animal (average no. recs about 3.5). I assume that
>the correct approach (also more robust?) for assessing the sig or otherwise
>of various fixed effects is to concurrently fit ide(animal) as a random
>effect in the model (rather than using SAS and excluding all random
>effects). Prior analyses indicate that many of the traits I am looking at
>are quite repeatable. It was suggested to me that I should also check the
>interaction year*ide(animal) in case this is a problem. I decided to do
>this, although I would have thought that a significant interaction here
>would simply co-incide with a trait of low repeatability? This gives rise
>to the question: if a trait is not repeatable do you need to in fact fit
>ide(animal) to check for sig. fixed effects. This is completely apart from
>the fact that if this were the result my trait of interest would also not
>appear heritable (which would cause a problem in itself :-)).
>
>Anyway, fitting this interaction (year*animal) can be a problem for some
>traits re convergence (which was not the effect I was trying to check for).
>What seems to happen is that this effect is bounded at zero if no
>qualifiers are used. Over-riding this with !GU means I can't get
>convergence at all (interaction estimate ranges from -ve to +ve). Is it
>safe to assume that this variance component is probably zero, or is it more
>correct to assume that there is not enough information in the data to do
>this test? Do I in fact need to do this test or am I wasting my time (and
>yours).
>
>Any suggestions?
>
>Thank you for your help!
>
>Cheers
>
>Kim
>Kim Bunter (M.Rur.Sc)
>PhD Student
>Animal Genetics and Breeding Unit
>University of New England
>Armidale, NSW, 2351
>AUSTRALIA
>
>Ph: (02) 6773 3788
>Fax: (02) 6773 3266
>email: kbunter@metz.une.edu.au
>--
>Asreml mailinglist archive: http://www.chiswick.anprod.csiro.au/lists/asreml
>
Kim Bunter (M.Rur.Sc)
PhD Student
Animal Genetics and Breeding Unit
University of New England
Armidale, NSW, 2351
AUSTRALIA
Ph: (02) 6773 3788
Fax: (02) 6773 3266
email: kbunter@metz.une.edu.au
--
Asreml mailinglist archive: http://www.chiswick.anprod.csiro.au/lists/asreml