Dear Ian,
>
>
> Arthur,
>
> Thank you for your reply last time, it has managed to point us in the
> right direction. But again we turn to you to find a simpler way of doing
> things. Our full model looks like
>
> gen year loc y.l e/t g.y g.l g.y.l e/g.t
>
> where e is coded to be y.l t is trial or management. Mark is interested
> in obtaining a pooled estimate of e/g.t and e/t.
For say 20 environments, e/t gives 20 t variances and
e/g.t gives 20 g.t variances
If you drop e/t from the model, e/g.t will increase to
include most [if not all] of the e/t variance.
I would have expected you would have fitted e [or y l y.l ] as fixed
effects but this is not clear above.
> We can do this by
> fitting everything except for e/g.t and using this as error but we need to
> subtract a pooled estimate of the trial variances as we are using
> mu+blup(gen) for each trial not the raw data.
I do not like the analysis of mu+blup(gen) obtained from the separate
analysis. I presume these are replicated experiments in which case
the means you are analysing should be mu+BLUE(gen) i.e. based on fixed
effects. You can then combine these in a weighted analysis.
> Is there a more explicit
> way of obtaining this pooled estimate of e/g.t through some coding in
> ASREML.
>
> Also, in the last email you sent me you setup an ASREML input file that
> optionally included G and R structures but I am not sure as to what some
> of them are.
>
> 6 1 6
> n1 !S2=s1
> n2 !S2=s2
> n3 !S2=s3
> n4 !S2=s4
> n5 !S2=s5
> n6 !S2=s6
>
> I am not sure what n1...n6 represent. Also as I don't have a site
> variance (I analysed each trial within each year location) I would omit
> the starting values for S2.
>
This relates to analysing the separate original trials. The n1 n2 ...
represent the number of observations in each trial and the s1 s2 represent the
trial variances. This is akin to the multisite analusis in
the ASREML manual (example 8.4) but it looks as if you are doing
something more like the grdc example (8.5).
> The next question I think I have asked before but I must not have worded
> it well. Is there a method for getting good starting values for the US
> structure or will they be guesstimates?
Look in the .res file. For a factor site.geno it will contain
empiracl estimates of of the US across site matrix after you
have fitted it simply [as independent effects].
> We have some results for this analysis from the old REML program into
> which the rep data was put. REML took a week to analyse only a very simple
> model but when we compared the results the answers were a little
> disturbing. We got a similar location variance component but the site and
> genotype vc were quite different. Do you have any ideas why this might
> be. Is it to do with putting the rep data into REML and spatial analysis
> means into ASREML or is there somethin more fundamental than that?
Since you mare using BLUPs as your data, you have already shrunk the effects.
YOu are shrinking them again - which is non sensible - hence you
should use fixed means for this second analysis, not BLUPs.
It is fine to use the BLUPS in an ordination context but not as data for
further analysis where they are being treated as random effects.
Arthur
>
>
> Thank you for your time.
>
> Ian
>
<><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><>
Arthur Gilmour PhD email: Arthur.Gilmour@agric.nsw.gov.au
Senior Research Scientist (Biometrics) fax: <61> 2 6391 3899
NSW Agriculture <61> 2 6391 3922
Orange Agricultural Institute telephone work: <61> 2 6391 3815
Forest Rd, ORANGE, 2800, AUSTRALIA home: <61> 2 6362 0046
ASREML is currently free by anonymous ftp from pub/aar on ftp.res.bbsrc.ac.uk
Point your web browser at ftp://ftp.res.bbsrc.ac.uk/pub/aar/
in the IACR-Rothamsted information system http://www.res.bbsrc.ac.uk/
To join the asreml discussion list, send the message
subscribe
to asreml-request@chiswick.anprod.CSIRO.au
The address for messages to the list is asreml@chiswick.anprod.CSIRO.au
<> <> <> <> <> <> <>
"Seek first the kingdom of God and His righteousness" Jesus; Matthew 6: 33
<><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><>