Date:         Wed, 31 Mar 2004 12:41:09 -0500
Reply-To:     Paul Kononoff <pjk2@MAPLE.UNH.EDU>
Sender:       "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From:         Paul Kononoff <pjk2@MAPLE.UNH.EDU>
Subject:      Missing data at the end of my curve

Hello folks,
I have sort of an unusual challenge. I am mining a data set which tests the
effects of a treatment on a response variable (back fat)  that employed
repeated measurements. The experimental unit was a beef steer.

The challenge is that once an animal reached a specific weight it was
slaughtered. I am trying to determine the effect of treatment on the rate of
back accumulation.

I have two problems:

1. Unfortunately the animals that grew the quickest, and with the steepest
slope, were killed the earliest.  Obviously I have more measurements on
animals that grew slower. The points that went unmeasured for slaughtered
animals were coded as missing data. Is this approach correct?

2. I also wanted to illustrate these results in a graph and have used an
LSMEANS statement to generate my points.  The resulting estimates at the
later time points seem very unreliable and biased, as they are heavily
influence by slower growing animals.  How should I determine estimates for
these time points? Maybe I can't

Any advice would help. For those of you interested I have included my code
below. Thanks in advance!
PJK

proc mixed data=UofS method=reml;
        class  PEN GT trt time;
        model BF = GT time GT*time /noint solution ddfm=satterth;
        random pen;
        repeated time/type=arh(1) sub=tag(pen);
  LSMEANS GT/diff;
  LSMEANS GT*time/slice=time diff;


  /**

       estimate 'CC v CT' GT  -1  1  0;
       estimate 'CC v TT' GT  -1  0  1;
       estimate 'CT v TT' GT  0  -1  1;

       estimate 'slopes CC v CT' GT*v_date  -1  1  0;
       estimate 'slopes CC v TT' GT*v_date  -1  0  1;
       estimate 'slopes CT v TT' GT*v_date  0  -1  1 ;


    estimate'slope overall' v_date 1;
       estimate 'slope CC' v_date 1 GT*v_date 1 0 0;
       estimate 'slope CT' v_date 1 GT*v_date 0 1 0;
       estimate 'slope TT' v_date 1 GT*v_date 0 0 1;
       run;


Paul Kononoff
Project Director
Ruminant Feed Analysis Consortium
University of New Hampshire
Durham, NH
PH: 603-862-1815