Date:         Mon, 19 May 2008 17:02:17 -0400
Reply-To:     Ingrid <ikf@HOTMAIL.COM>
Sender:       "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From:         Ingrid <ikf@HOTMAIL.COM>
Subject:      GLIMMIX and model fitting

I am trying to work with glimmix to do a logistic regression.  I was
planning on using both a random effect (id number) as well as a repeated
measures effect (time)...This does not converge with the smaller version
of the dataset.  However, I can use either the random or the repeated.  I
have good reasons to belive that each one might be better than the other -
but was looking for a statistical basis for my decision.  I know that I
cannot use the pseudo-likelihood.  However, the generalized chi-square/df
is not different, although the gen chi-square itself is different.

I know others have asked the same question, but no-one seems to have an
answer...how do I decide which effect is 'better'?

Ingrid