Date:         Sun, 29 Mar 1998 21:19:27 -0500
Reply-To:     Terry Hyslop <Terry.Hyslop@MAIL.TJU.EDU>
Sender:       "SAS(r) Discussion" <SAS-L@UGA.CC.UGA.EDU>
From:         Terry Hyslop <Terry.Hyslop@MAIL.TJU.EDU>
Organization: Thomas Jefferson University
Subject:      Re: PROC MIXED
Content-Type: text/plain; charset=us-ascii

Tra wrote:
>
 
>      I would  like some help in obtaining covariance matrices from proc
>      mixed.
>
>      I am fitting a random coefficients regression model.  Only two of the
>      coefficients have a significant random element.  I need to obtain the
>      variance-covariance matrix for the coefficients
>
>      (1)  for a particular class within the dataset
>
>      (2)  for a new class, assumed to be in the same population.
>
>
>      I cannot see how to get at case (2), except by adding a dummy class to
>      the dataset with missing data for z1-z5 and y.
>
>      The matrices produced by the V=28 and VI=28 options to the random
>      statement have 9 rows and 9 columns.  I cannot work out what these
>      matrices are, and the documentation does not help.
>
. . .
 
Tim,
 
The V matrix you are printing is generated by the random
statement in your Proc Mixed code:
 
V= ZGZ' + R
 
where Z is the design matrix whose columns are based on you random
statment,
G is the variance-covariance matrix of the random effects,
and R is the residuals.  Since you have not added any structure to the
residuals
through a repeated statment, the R matrix is (sigma**2)*I.
 
If you just want the variance-covariance matrix, then add the G option
to your
random statement.
 
The BLUP (Best Linear Unbiased Predictor) of the variance-covariance
parameters for an
individual subject (class in your case) can be obtained through the
SOLUTION option on the
random statement.
 
Hope this helps,
 
Terry Hyslop
Senior Biostatistician
Thomas Jefferson University