Re: Region by Environment models...

From: Scott Chapman <Scott.Chapman_at_CSIRO.AU>
Date: Fri, 29 Feb 2008 08:58:15 +1000

Thanks Arthur, especially for the suggestion about:
corh(Reg):Gen + corh(Env):Gen
I wasnt sure about the implications for nesting this.
Because the input data in this case is BLUEs plus weights, I do already
account for the heterogenous error variance. If I was using the raw
data, then I would use rcov ~ at(Env):ar1(Row):ar1(Col) for example.


        From: ASReml users discussion group
[mailto:ASREML-L_at_AGRIC.NSW.GOV.AU] On Behalf Of
        Sent: Friday, 29 February 2008 8:29 AM
        Subject: Re: Region by Environment models...

        Dear Scott,
        Again I will reply without the detailed ASReml-R syntax
        The model I used in the previous email was pretty a basic
compound symmetry model.
        a.asr <- asreml(fixed = gyp ~ Env,
                random = ~Genotype + Region:Genotype +
                rcov = ~units)
        ** The RANDOM model can equivalently be written as
                random = ~Genotype + Region:Genotype + Env:Genotype,
        I would like to use better corh or fa structures in this
analysis. Ideally, these would be fitted to the entire dataset and then
the data predicted by Region somehow...
        For example, I know that a corh model fits much better to the
entire set of environments in this dataset:
        a.asr <- asreml(fixed = gyp ~ Env,
                random = ~corh(Env):Genotype,
                rcov = ~units)
        ** The extention from the Simple model is
                 random = ~corh(Reg):Genotype + corh(Env):Genotype,
         BUT it does not make sense to allow heterogeneous genetic
variance without allowing heterogeneous error variance.
        Given a likely 10fold + range in phenotypic variance, all levels
need to be heterogeneous.
        Prediction of Reg x Geno is just as for the simpler model.
        Your following suggestions are not appropriate.
        However, I would like to be able to predict at each region, so
can I somehow nest the corh model within region, or will I have to run
these separately?
        a.asr <- asreml(fixed = gyp ~ Env,
                random = ~diag(Region):corh(Env):Genotype,
                rcov = ~units)
        or perhaps:
         a.asr <- asreml(fixed = gyp ~ Env,
                random = ~at(Region, 1):corh(Env):Genotype + at(Region,
                rcov = ~units)
        The problem with these models appears to be that I would need to
fit the corh only to the Envs that occur within each Region. Does that
mean I need to have two new Env factors where they are declared NA for
the regions where they do not occur?
        May Jesus Christ be gracious to you in 2008,
        Arthur Gilmour, His servant .
        Mixed model regression mapping for QTL detection in experimental
crosses. Computational Statistics and Data Analysis 51:3749-3764 at
        Personal website:
        Skype: arthur.gilmour,
        Principal Research Scientist (Biometrics)
        NSW Department of Primary Industries
        Orange Agricultural Institute, Forest Rd, ORANGE, 2800,
        fax: 02 6391 3899; 02 6391 3922 Australia
        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,
        Archives are at
        Proposed travel:
        Argentina 16-19 September Biometrics 
Received on Mon Mar 01 2008 - 08:58:15 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.)