Re: how do I predict across regions and deal with aliasing and nesting in ASREML-R?

From: <arthur.gilmour_at_DPI.NSW.GOV.AU>
Date: Tue, 26 Feb 2008 09:45:57 +1100

Dear Scott,

re: On the bottom of page 65 of the ASREML-R manual (release 2.0, Feb
2007), there is this text:
"The present argument enables the construction of means by averaging only
the estimable cells of the hyper-table. It is reguarly used for nested
factors, for example locations nested in regions."
Unfortunately, there is no example....
I have some across region analyses of crop genotype (G) data with BLUEs
and weights as input. I wish to estimate the hyper-table of Region x
 Has anyone got some examples of models/predictions using the present
argument to do this? Whatever model I use in ASREML-R I usually end up
with 'aliasing' and therefore a set of NAs for my predictions. I can
override the aliasing, but am afraid that I might be overfitting in any
case.... Some of the models I've tried are:
Fixed = Env, random = ~ G + Reg:G + Reg:Env:G
Fixed = Env, random = ~ corh(Reg):G
Fixed = Env, random = ~ diag(Reg):G + diag(Env):G Am not sure if this
one is legit - am assuming that the regional interaction will be taken out
first... Unless there is some way to nest it within regions...?

I can answer this with respect to ASReml and hope you can apply the answer
to ASReml-R syntax.
The funcrionality should be equivalent.

First, concerning the models.

Reg:Env is equiavalent to Env if the Environments are coded uniguely
across regions. i.e. if Envoronments are place names, they
will be unique across regions. It is of course possible to code
Environments as 1, 2, 3 etc within regions in which case the levels
of Region (say if environments represented Early/Late or Irrigated/Dry
rather than places)

Second, Aliassing in prediction relates to fixed effects, rather than
random effects,
so assuming you have shown the whole model above, aiassing will mean that
some Environments have no data.

However, in that case !PRESENT Region Environment should resolve the
More likely you have fitted some other fixed interaction which has a
missing cell so that it cannot be averaged.

Consider the example
Region 1 1 1 1 2 2
Expt 1 2 3 4 5 6

This is the situation envisaged by the User Guide comment.
When calculating Regional means,
  for region 1, we want to average over environments 1 2 3 4
for region 2, over environments 5 6

PREDICT G Reg !present Reg Exp

will do that. We do not want to include G in the !PRESENT list,
assuming some G may not appear in some experiments,
 because that would mix up Env effects into the comparison of Genotypes
within a Region.

I trust this helps.

the genotype effects a

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, 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,
Archives are at

Proposed travel:
Argentina 16-19 September Biometrics 
Received on Fri Feb 26 2008 - 09:45:57 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.)