Date:         Wed, 22 Mar 2006 14:16:03 -0800
Reply-To:     Dale McLerran <stringplayer_2@YAHOO.COM>
Sender:       "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From:         Dale McLerran <stringplayer_2@YAHOO.COM>
Subject:      Re: Specifying a residual correlation matrix
In-Reply-To:  <1143054180.259751.124190@t31g2000cwb.googlegroups.com>
Content-Type: text/plain; charset=iso-8859-1

--- Eson <ulf.emanuelson@KV.SLU.SE> wrote:

> Hi all,
>
> I wonder if it is possible to specify a tailor-made residual
> correlation matrix (R) to be used in a mixed model analysis (PROC
> MIXED
> or PROC GLIMMIX)?
>
> The reason for the question is that we have observations on a number
> of
> individuals (one obs per individual) and we are interested in
> estimating the effects of some (fixed) explanatory variables on the
> outcome. However, we know that these individuals are relatives and
> would like to account for this similarity. The idea was then to force
> SAS to use an R-matrix that we have defined and that show the
> relationship between them, since we believe that e is not
> ND(0,I*sigma_e), but rather ND(0,R*sigma_e). I believe this is pretty
> straight forward MME, but I am not sure that SAS can handle it.
>
> All and any ideas are welcome!
>
> Ulf
>

Ulf,

You don't say whether there are constraints on the covariance
structure indicated by R.  The MIXED procedure can estimate many
covariance structures and chances are that the covariance
structure that you would assume is among those that are available
in the MIXED procedure.  If you were to state a little bit more
about your assumptions regarding R, then we can offer a more
definitive statement.

Occassionally one encounters a covariance structure which cannot
be effectively estimated employing the MIXED procedure.  In such
cases, one can often estimate the model employing the procedure
NLMIXED.  Whether the model can be estimated employing NLMIXED may
depend on the number of random effects one must assume.  When you
have R-side (residual) random effects, then if you are to employ
NLMIXED to estimate your model, you must specify a separate random
effect for each individual within a family group.  If a family
group is quite large (say, 15 or more), then you may run into
problems using NLMIXED.  I would note that NLMIXED can become
quite inefficient as the number of random effects gets large.

Bottom line: we can't really give you any definitive statement
unless you tell us more about your assumptions about R.

Dale


---------------------------------------
Dale McLerran
Fred Hutchinson Cancer Research Center
mailto: dmclerra@NO_SPAMfhcrc.org
Ph:  (206) 667-2926
Fax: (206) 667-5977
---------------------------------------

__________________________________________________
Do You Yahoo!?
Tired of spam?  Yahoo! Mail has the best spam protection around
http://mail.yahoo.com