Date: Mon, 20 Mar 2006 20:27:52 -0500
Reply-To: Anna M <annamaine74@HOTMAIL.COM>
Sender: "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From: Anna M <annamaine74@HOTMAIL.COM>
Subject: Re: Hierarchical model with MIXED or GEE in GENMOD, GLIMMIX,
Content-Type: text/plain; format=flowed
Unfortunately, I don't have data quite yet so cannot test the distribution
of my dependent variable. I'm perfectly willing to jump over to NLMIXED or
something else if needed though. But will NLMIXED or GLIMMIX or GEE in
GENMOD choke on my sample sizes of 470,000+ students.
I also recognize that this type of analysis is fraught with challenges and
questionable results but some sort of analysis must be done. Don't worry
everyone, absolutely no teachers' jobs are on the line in any way whatsoever
with this analysis.
Regarding the comment: Whether to use nlmixed depends on how much prior
knowledge you have about the functional form of the model. So some
exploratory analysis might be very helpful. My response: This comment was
very interesting to me, what would I be looking for with an exploratory
analysis? In other words, what do you mean by functional form of the model?
I am specifically interested in the set of teachers/schools/districts that I
have, so I agree that I don't think I have any random effects for these
particular terms. But I need a mixed model so I can treat student as a
random effect because I have multiple observations (3 yrs worth) on a
student, right? If I only had fixed effects, wouldn't I just be able to do
the whole analysis in GENMOD? (I'm more familiar with GENMOD.)
David, how would you build the model: teachers whose classes have the
largest improvements over the prior-year results? This is one kind of
question of interest to me but I'm just not sure how to specify it.
I think my DV is testscore but is it somehow change in test score instead?
My IVs are:
economicinfo (free lunch, reduced lunch, not eligible)
teacher nested within school which is nested within district (or something
I guess my struggle is that I'm just not seeing how a model looking at
teachers whose students have the highest test scores differs from a model
for teachers whose students have the largest improvements in test scores.
Also how to write a triple nesting of teacher within school within district
has me stumped. I'm not sure my glimmix with nesting in a prior email was
Also, anyone that has seen a book that explains nested models in a
straightforward way, please let me know.
Thanks so much for your help here everyone.
Donít just search. Find. Check out the new MSN Search!