Re: Predict GxE for XFA - responses

Dear All,

Another helpful response I received was...

> you need to be fitting
> "-1+env" as a fixed effect instead of "mu"
> this will give you a mean for each env and will ensure the blups at
> each env are shrunk towards the env mean rather than an overall mean.
> you are correct, it is perfectly fine to add the mu (or in this case
> the env blue) to the gen blups with respect to adding back the
environmental
> effect

and another
> BLUE + BLUP = BLUP, and fit envs as fixed.

In ASREML fitting "-1" is equivalent to not fitting "mu". I believe the
'-1' to be for SAMM syntax. (Certainly fitting "-1 +env", and "env" gave
the same .sln file)

I found this didn't affect the results from the predict env.gen
statement at all but I could go to the .sln file and add up the bits I
wanted, i.e coeff(e1) + coeff(e1.g1) BLUPs.
In fact, I then realise I could have done this in the first place:
BLUP(e1,g1) = mu + coeff(e1)+ coeff(e1.g1)

Moral of story: as always know what your model is telling you and read
the .sln file

Thanks for your patience and assistance.
Ky

-----Original Message-----
From: ASReml users discussion group [mailto:ASREML-L_at_AGRIC.NSW.GOV.AU]
On Behalf Of Dr Ian DeLacy
Sent: Wednesday, 9 August 2006 8:11 PM
To: ASREML-L_at_AGRIC.NSW.GOV.AU
Subject: Re: Predict GxE for XFA

Ky
The mu is not the mean but a multiple regression constant and is almost
never the mean. You can only make mu the mean if you set all the fixed
effects to sum to zero a very troublsome thing in what you are really
meaning to do. WHAT IS THE AVERAGE HEIGHT OF WHEAT OR THAT WHEAT LINE? I
totally avoid the problem by fitting a model such as
... g gy gl gyl if I want overall cfs among entries over all data
or
... gl gyl if I want to cf the performance of entries in each location
averaged over years or ... gy gyl if I want to cf the performance of
entries over years averaged over locations. I think it is best to live
with all these values summing to zero then trying to rescale them If you
want to give breeders or others a feeling for the values (height being a
case where they are interested) predict overall mean or site means of
year means using the BLUEs. Cheers Ian

Quoting Ky Mathews <Ky.Mathews_at_CSIRO.AU>:

> Dear ASREML-colleagues,
>
> I'm fitting an factor analytic (XFA2) model to a multi-environment
> trial for wheat data.
>
> I want to predict the GxE BLUPs an use the following code
>
> predict env.gen !only xfa(env,2).gen
>
> This gives me BLUP estimates but we would like them to be on the same
> scale as the measurements, i.e. I obtain mu from the .sln file and add

> it to the BLUP estimates.Thus, we're not changing any ranking cos
> we're simply adding the same constant to all BLUPs.
>
> Is this correct? (I thought so!)
>
> But... in recent analyses of height data I see that the predicted
> values for some envs are well below what was actually observed. I know

> that BLUPs get shrunk back the mean (mu, I assume?), but is there a
> way to adding back in the environment main effect? I suspect that this

> may not be a sensible thing to do as environments are fixed, so we'd
> have a BLUE + BLUP which does not equal a BLUP!
>
> Any suggestions/advice would be much appreciated.
>
> Thanks
> Ky
>
> Dr Ky L Mathews
> CSIRO Plant Industry
> Queensland Bioscience Precinct
> 306 Carmody Road
> St Lucia QLD 4067 Australia
> Ph: +61 7 3214 2255
> Fax: + 61 7 3214 2920
> Email: ky.mathews_at_csiro.au
> "It is not the strongest of species that survive, nor the most
> intelligent, but the one most responsive to change." - Charles Darwin
>
>
>
>
>
>
>
>

Dr Ian DeLacy
Lecturer in Genetics and Plant Breeding
School of Land and Food Sciences
University of Queensland
Brisbane, 4072
Australia
Phone +61 (7) 3365 2833
E Mail i.delacy_at_uq.edu.au
Received on Tue Aug 10 2006 - 09:59:43 EST