|
Thank you for helping me solve the "infinite likelihood" problem in running
Proc Mixed. I succeeded in running the analysis.
Now I am faced with the challenge of interpreting the analysis result in
light of my data:
Here is part of the effect estimates I don't know how to explain (I only
listed a couple of the significant estimates that give me trouble):
Effect B C Time Estimate
----- -- -- ---- --------
Intercept 59.8
Time 1 -6.5
Time 2 0
C 1 -12.0
C 2 0
B*Time 1 1 7.2
B*Time 1 2 0
B*Time 2 1 5.7
B*Time 2 2 0
B*Time 3 1 0
B*Time 3 2 0
---------------------------------------------
I am confused about 2 types of estimates:
(a) the intercept - Is it the overall mean estimate or a mean estimate of
some reference sub-group? Is it an adjusted mean by the two covariates, D
and E, in the model?
(b) the "0" estimates - I know they are the reference estimates that the
other effects within the group are compared to. But it gets to be very
confusing to explain these estimates to others (including me), especially
those involving interactions.
Please help.
Susie Chung Li
-----Original Message-----
From: SAS(r) Discussion [mailto:SAS-L@LISTSERV.UGA.EDU] On Behalf Of Dale
McLerran
Sent: Tuesday, April 03, 2007 4:23 PM
To: SAS-L@LISTSERV.UGA.EDU
Subject: Re: Error from running Proc Mixed for Repeated Measures Analysis of
covariance
--- Chung Li <cli2@RDG.BOEHRINGER-INGELHEIM.COM> wrote:
> Hello,
>
> I have a large repeated measures analysis of covariance model that
> combines a
> two factor experiment (B - 4 levels, and C - 3 levels) with two
> independent
> covariantes (D and E). The "Time" variable has only 2 values: 1 and
> 2.
>
> There are 8,649 subjects in the experiment, and a total of 17,298
> data points
> to be analyzed. I ran the following SAS code:
>
> proc mixed data=test;
> class B C ID Time;
> model Y = Time|B|C D E B*D B*E C*D C*E /noint;
> repeated Time / type=cs subject=ID;
> run;
>
> Here is the error message I got:
>
> Iteration History
> Iteration Evaluations -2 Res Log Like
> 0 1 139413.55872390
>
> WARNING: Stopped because of infinite likelihood.
>
> Any idea wheat I could have done wrong?
>
>
> Susie C Li
>
Susie,
I don't see a problem with your code. The error message that you
see can be produced if you have some subjects who have improperly
coded data such that there are two records with the same ID and
same time value. Have you checked to determine that there are no
data construction problems of that sort? You can do a quick check
of this issue by running the code
proc sort data=test;
by ID time;
run;
data _null_;
retain error 0;
set test;
by ID time;
if (first.time+last.time)^=2 then do;
error+1;
if error=1 then
put "ERROR in data construction. Two or more records" /
"have the same ID and time value. These data must" /
"be fixed before running a mixed model with a" /
"REPEATED TIME / subject=ID; statement. Data in" /
"error have the following ID and TIME values:";
put ID= TIME=;
end;
run;
Sometimes, the MIXED procedure does run into estimation problems
like this for no apparent reason. I have seen it myself and not
been able to identify just why MIXED fails. Often, I have been able
to get around the problem through some reparameterization of the
model. There are two eparameterizations of the model which should
produce exactly equivalent results for the mean and variance
structure as the model you have coded.
First, you have specified the option NOINT. Why? Removing the
intercept through specification of the NOINT option does nothing to
change the mean model when you have categorical variables included
in your fixed effect design matrix. All that you really accomplish
is that you make it more difficult to construct correct effect
tests. So, the first thing that I would do is remove the NOINT
option and try fitting the model again.
Second, I would note that when you have only two time values, there
are at least two other covariance structure which are identical to
the TYPE=CS covariance structure. These are TOEP and AR(1). I
have had success getting around the problem you report simply by
changing the covariance structure from CS to TOEP or AR(1). You
might try that.
HTH,
Dale
---------------------------------------
Dale McLerran
Fred Hutchinson Cancer Research Center
mailto: dmclerra@NO_SPAMfhcrc.org
Ph: (206) 667-2926
Fax: (206) 667-5977
---------------------------------------
_____________________________________________________________________________
_______
It's here! Your new message!
Get new email alerts with the free Yahoo! Toolbar.
http://tools.search.yahoo.com/toolbar/features/mail/
| 
