Date:         Fri, 11 Mar 2011 21:21:42 -0500
Reply-To:     R B <ryan.andrew.black@GMAIL.COM>
Sender:       "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From:         R B <ryan.andrew.black@GMAIL.COM>
Subject:      Estimating Difference in Centroids in Multivariate Linear Mixed
              Model
Content-Type: text/plain; charset=ISO-8859-1

Hi,

Dale has shown how to contruct a multivariate test employing the
CONTRAST statement in MIXED. CONTRAST statements, however, do not
provide actual estimates (e.g. difference in centroids between
groups). Is there any way to produce the same multivariate test
employing the ESTIMATE statement in order to obtain the difference in
centroids?

I provide a little simulation along with MIXED code below to help
concretize my question.

Thanks,

Ryan
--

data test;
 seed = 987624;

  do subject = 1 to 1000;
  group = ranbin(seed,1,0.50);
  random_eff = sqrt(0.3)*rannor(seed);

     do y_indic = 1 to 3;
     err = rannor(seed);

     y = 1.0*(y_indic=1)
         + 2.0*(y_indic=2)
         + 3.0*(y_indic=3)
         + 1.0*(y_indic=1)*(group=0)
         + 2.0*(y_indic=2)*(group=0)
         + 3.0*(y_indic=3)*(group=0)
         + random_eff
         + err;

     output;
   end;
 end;
run;

proc mixed data = test;
class y_indic group;
model y = y_indic y_indic*group / s noint;
repeated y_indic / subject = subject type = un;
contrast 'multivariate test'
         y_indic*group -1 1  0 0  0 0,
         y_indic*group  0 0 -1 1  0 0,
         y_indic*group  0 0  0 0 -1 1;
run;