Date:         Tue, 13 Jul 2010 08:24:10 -0500
Reply-To:     Robin R High <rhigh@UNMC.EDU>
Sender:       "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From:         Robin R High <rhigh@UNMC.EDU>
Subject:      Re: PROC MIXED - why would AR(1) model be exactly like CS model?
Comments: To: peterflomconsulting@mindspring.com
In-Reply-To:  <001f01cb228b$1789b390$469d1ab0$@com>
Content-Type: text/plain; charset="US-ASCII"

Peter,

Typically with longitudinal data, one treats the covariance matrix as
R-Side (residual) and applies the REPEATED statement with MIXED for these
situations. Perhaps one is now more easily led astray since GLIMMIX has
only a RANDOM statement with the residual option to be added.


 repeated pid/type = cs subject = pid R ;


 repeated pid/type = ar(1) subject = pid R;


will likely give you different results.  You get the same results with
RANDOM because there is only 1 random factor.

Robin High
UNMC





From:
Peter Flom <peterflomconsulting@MINDSPRING.COM>
To:
SAS-L@LISTSERV.UGA.EDU
Date:
07/13/2010 08:02 AM
Subject:
PROC MIXED - why would AR(1) model be exactly like CS model?
Sent by:
"SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>



Hello



I am working with a longitudinal data set.  Six waves of data on 37
people.
With such data, I usually try CS and AR(1) and sometimes other structures.
This time, I get identical results from the two structures (same parameter
estimates, same fit indexes etc) and the estimate for the covariance for
AR(1) is 0 (although it is not 0 for CS.



PID = ID number

ITT is a class variable for treated vs. control

Anxdep is the dependent variable, a measure of anxiety and depression

Wave is the wave number (1 to 6).



Code for CS:

<<<<<<<<<<

title 'ITT analysis';

title2 'Factor 1: AnxietyDepression';

title3 'Random intercepts only';

title4 'CS covariances';

ods html;

ods graphics on;

proc mixed data = schooler.long plots=studentpanel (marginal);

 class pid itt;

 model anxdep = itt wave itt*wave/solution residual;

 random pid/type = cs subject = pid;

run;

ods graphics off;

ods html close;

>>>>>>>>>>>>>



Code for AR(1)



<<<<<<<<<

title4 'AR(1) covariances';

ods html;

ods graphics on;

proc mixed data = schooler.long plots=studentpanel (marginal);

 class pid itt;

 model anxdep = itt wave itt*wave/solution residual;

 random pid/type = ar(1) subject = pid;

run;

ods graphics off;

ods html close;

>>>>>>>>>>>



What am I missing here?



Thanks



Peter



Peter Flom PhD.

Peter Flom Consulting LLC

5 Penn Plaza, Ste 2342

NY NY 10001

www.statisticalanalysisconsulting.com

www.IAmLearningDisabled.com