Date:         Tue, 16 Mar 1999 10:18:42 -0500
Reply-To:     Tim Phelps <phelpst@PSU.EDU>
Sender:       "SAS(r) Discussion" <SAS-L@UGA.CC.UGA.EDU>
From:         Tim Phelps <phelpst@PSU.EDU>
Organization: Penn State University
Subject:      Hierarchial (nested) random effects model
Content-Type: text/plain; charset=us-ascii

Hello,
    I have been trying, without much success, to analyze a hierarchial
(nested) random effects model using PROC GLM and PROC VARCOMP.  My
effects are all random and look something like this:  X, D, D(C), C(B),
B(A).  I need to test each variable based on EMS and variance
components.  The RANDOM statement under PROC GLM does a good job of this
up to certain point.  Basically, I can successfully analyze X, D, X*D,
D(C), C using this statement.  However, when I add the nested effect C
within B (i.e. C(B)) I get nothing.  Logistically, this makes sense
since the variability has already been explained by D(C) and C.  Is
there someway I can isolate the nested variables to analyze them
seperately  The problem is that EMS for each nested variable should be
somewhat of an additive, so they all need to be included in the model
statement (or so it seems).  I'd sure hate to do this by hand.
 
    Here's an example of a program that works and one that does not:
proc glm data=XXXX.XXXX;
    class X D C B A;                                |
    .                                                          |
    .                                                          |
works fine
    random X D(C) X*D(C) C / test;        |
 
 
    random X D(C) X*D(C) C C(B) B B(A) A / test; |    doesn't work
 
Any suggestion or ideas would be greatly appreciated.
 
Thanks,
Tim