Date:         Mon, 16 Feb 2009 16:09:18 -0600
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 - Rsq
Comments: To: nuria <nchapinal@YAHOO.COM>
In-Reply-To:  <60f396d3-8cf9-48a9-b319-b339892c055e@v5g2000pre.googlegroups.com>
Content-Type: text/plain; charset="US-ASCII"

Nuria,

If you have _independent_ data, PROC GLM works virtually the same as PROC
MIXED with method=type3, so could get R-square the usual way. If you have
random effects or repeated measures, r-square doesn't exist. There are two
ways I know about to compute pseudo-rsquare values based on ratios of
variance components as described in "Applied Longitudinal Data" by Singer
and Willett or "Multilevel Analysis" by Snijders and Bosker (Chpt. 7).
However, they can produce negative estimates and they do not necessarily
agree with one another; my experience with any type of rsquare is "mixed"
and difficult to explain, so usually don't find it worth the trouble to
pursue -- focus on graphical displays and explaining the sources of
variation.

Robin High
UNMC





nuria <nchapinal@YAHOO.COM>
Sent by: "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
02/13/2009 01:14 PM
Please respond to
nuria <nchapinal@YAHOO.COM>


To
SAS-L@LISTSERV.UGA.EDU
cc

Subject
Proc mixed - Rsq






Hi,

When I run Proc GLM I get the Rsq for the model by default. How can I
ask for it in Proc MIXED?
Thanks!