Date: Tue, 30 Jun 2009 08:23:19 -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: Matched Triplets - Glimmix
In-Reply-To: <d657ff4b-9c9a-4150-84aa-8fecd782ee46@j14g2000vbp.googlegroups.com>
Content-Type: text/plain; charset="US-ASCII"
Jeff,
The basic features look OK, though with one random effect and for groupid
clusters of size 3, I'd work with method=laplace or method=quad (rather
than the default) (also, would prefer ddfm=contain (the default0 rather
than bw, but would otherwise also recommend ddfm=kr, though not available
with method=quad). Also if you are interested in differences in the
agetype variable, I'd add the diff and oddsratio options to the LSMEANS
statement (remove the latter from MODEL) and add a multiple comparison
adjustment to the confidence intervals:
proc glimmix data = final method=quad; *or perhaps method=laplace;
class agetype groupid;
model sponsored08_binary (desc) = agetype / dist=binary link=logit
solution ;
random intercept / subject =groupid;
lsmeans agetype / cl ilink diff oddsratio adjust=simulate(seed=39283
nsamp=50000);
run;
Robin High
UNMC
Jeff <jeffrey.m.allard@GMAIL.COM>
Sent by: "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
06/30/2009 07:15 AM
Please respond to
Jeff <jeffrey.m.allard@GMAIL.COM>
To
SAS-L@LISTSERV.UGA.EDU
cc
Subject
Matched Triplets - Glimmix
Hey Guys-
Looking for some validation for my first foray into the world of Proc
Glimmix (and Generalized Linear Mixed Models).
The Setup:
*I have individuals from three age groups (generations). I am looking
at a very simple situation where I want to evaluate the generations
for differences on a couple different measures (some binary, some
counts, 1 continuous). I have excluded all individuals that did not
have all of 2008 to accumulate these events.
*One big difference between individuals is the length of their tenure.
How long they have been members will likely effect the dependent
variables I'm interested in.
*So, I have taken the age group I am the most interested in, and
matched each individual to a member of the other two age groups on the
basis of tenure. The matches (measured in number of days prior to
1/1/2008 is nearly perfect for each.
*I now have matched triplets, where each member of each triplet is
from one of the three age groups.
Proc Glimmix:
*I have seen a couple examples of using Glimmix with matched pairs
(Peter Floms paper on dyads for example) but never quite like what I
am looking to do.
*Thoughts on this? Here is the simple Glimmix code:
proc glimmix data = final;
class agetype groupid;
model sponsored08_binary (desc) = agetype / dist=binary link=logit
ddfm=bw solution oddsratio;
random intercept / subject =groupid;
lsmeans agetype / cl ilink;
run;
Has my first Glimmix experience been a boom or bust?
Thanks!
Jeff
