Date: Tue, 21 Mar 2006 11:03:49 -0500
Reply-To: Jing Quan <JXQUAN@SALISBURY.EDU>
Sender: "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From: Jing Quan <JXQUAN@SALISBURY.EDU>
Subject: Re: HLM?
Content-Type: text/plain; charset=US-ASCII
This is a follow-up to my previous post. What I try to accomplish here
is to see if variable z should be included in the estimation, which may
be a moderator. I think I can use the t-ratio to test the significance
of c2, but some literature also suggests to test whether the increase in
adjusted R-square after including z is significant and suggested to use
HLM. But I don't see how HLM can accomplish this.
Thank you Peter for your email. I'm new to this list and did not know
exactly what to ask. I hope this one would clarify what I try to do
Jing "Jim" Quan, Ph.D.
Dept of Info and Decision Sciences
>>> "Peter Flom" <Flom@ndri.org> 2006-3-21 10:51 >>>
>>> Jing Quan <JXQUAN@SALISBURY.EDU> 3/21/2006 10:26:01 am >>> wrote
I try to test the following two models:
y = a1 + b1x + e1
y = a2 + b2x + c2z + e2
to see if the R-square for the 2nd equation is significantly larger
than that for the 1st one. If so, z should be included.
HLM is supposed to do this. But after reading all the literature so
far, none of them seems to address this question. They all talk about
level 1 vs. level 2 and fixed vs. random.
I'm not at all sure why you are using anything from HLM.
Why not just PROC GLM? or REG?
Of course, since you don't say why you are doing what you are doing,
any advice that anyone gives could be totally wrong. But, based on what
you've said, GLM or REG should do it.
why don't you write back to SAS-L (not just to me) saying what you are
trying to do, and then maybe someone can provide more cogent advice
Peter L. Flom, PhD
Assistant Director, Statistics and Data Analysis Core
Center for Drug Use and HIV Research
National Development and Research Institutes
71 W. 23rd St
New York, NY 10010
(212) 845-4485 (voice)
(917) 438-0894 (fax)