Date:         Wed, 15 Aug 2007 08:47:23 -0700
Reply-To:     Paul Miller <pjmiller_57@YAHOO.COM>
Sender:       "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From:         Paul Miller <pjmiller_57@YAHOO.COM>
Subject:      PROC GENMOD: Specifying the Working Correlation Matrix for
              Repeated Measures Data
In-Reply-To:  <200708150500.l7F07rK9014872@mailgw.cc.uga.edu>
Content-Type: text/plain; charset=iso-8859-1

Hi David,

  Thanks for your reply.

  Earlier, I looked at some correlations and covariances and found what looks like a mix of autoregressive and exchangeable structures. There was a tendency for the association of early and subsequent tests to die down, but only to a point. So, for example, after say the 6th or 7th test, associations between the first test and each subsequent test (e.g., test 8, 9, and 10) tended to be pretty constant.

  I’m being asked to run some models with a lot of predictors and then to pare things down until I have something more parsimonious. As such, it looks like I need to select some kind of structure. And my sense if that an autoregressive structure makes the most sense, even if it doesn’t perfectly describe what appears to be happening in the data.

  Another person who responded (without posting to the list) suggested that an autoregressive structure might only be suitable if the lipids measurements were equally spaced. So I was wondering what your opinion is on that.

  My lipid measurements are not necessarily equally spaced. My sense from looking at the data is that physicians typically monitor patients who do not have elevated lipids at regular intervals but then might start ordering tests more frequently once a problem arises.

  Thanks,

  Paul


---------------------------------
Yahoo! oneSearch: Finally,  mobile search that gives answers, not web links.