I'm glad I met you last week and you have arrived home all fired up.
I think you are in fairly uncharted waters when it comes
to many highly correlated traits. You could divide your
traits into 7 groups (5 in each) and then try and
analyse each group with each other. I would chose the 5
in each set to be highly correlated among themselves.
If you find 2 traits have genetic correlation > 1, then
obviously you will have trouble fitting them together
with other traits because of lack of positive definiteness.
using the !GP option may succeed in fixing the correlation to .999.
What is really needed in this case is consider linear functions of
traits (as in Principal components). Then you would consider
To this end, you might try the Factor Analytic model
on a group of highly related traits so as to define a Factor
which picks up the dominant variation from the set of traits.
Then you could consider analysing linear functions of the
original traits so that all the information is included in the analysis.
I hope this gives you a few ideas,
> From email@example.com.CSIRO.AU Tue Jan 20 10:27:09 1998
> Date: Tue, 20 Jan 1998 09:43:44 +1030 (CDT)
> From: Jane Hill <firstname.lastname@example.org>
> Subject: Re: Multi-variate Correlations
> To: Arthur Gilmour <email@example.com>
> Cc: firstname.lastname@example.org.CSIRO.AU
> Mime-Version: 1.0
> To all ASREML users,
> I am currently doing my phD at The University of
> Adelaide. My project involves analysing fleece and skin measurements
> taken from the South Australian Turretfield Research Flock.
> We have a situation in which there are approximately 15 skin traits and
> 20 fleece traits, with which we want to calculate between trait
> correlation values.
> We have already conducted bivariate analysis amongst these traits, and we
> are wanting now to calculate some multivariate correlation values. The
> memory on our computer restricts us to 11 traits per run, and there are
> certain traits that cannot be grouped together because it appears as if
> LogL will not converge when traits that are highly correlated are grouped
> We would like some comments and advice, based on other users experience,
> on how to logically group our traits. Should we aim to get as many
> traits in a run as possible, or should we make more groups with smaller
> numbers? Is there a point when you stop increasing the number of traits
> included in a particular run? Are there certain traits that should be
> grouped together to obtain the most accurate correlation values?
> These are a few of our questions, can users reply with details on what
> method they have used to calculate multivariate correlation values
> amongst large data sets.
> Jane Hill
> The University of Adelaide
> South Australia.
Arthur Gilmour PhD email: Arthur.Gilmour@agric.nsw.gov.au
Senior Research Scientist (Biometrics) fax: <61> 2 6391 3899
NSW Agriculture telephone work: <61> 2 6391 3815
Orange Agricultural Institute home: <61> 2 6362 0046
Forest Rd, ORANGE, 2800, AUSTRALIA
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
The address for messages to the list is email@example.com.CSIRO.au
>>>>>>>>>>>> ASREML WORKSHOP sheduled 10-12 March: $300 <<<<<<<<<<<<<
But He answered and said, "It is written, `Man shall not live by bread alone,
but by every word that proceeds from the mouth of God' " Matthew 4:4