Date: Thu, 12 Jun 2008 13:24:02 -0400
Reply-To: Hideko Bassett <hhamada@gmu.edu>
Sender: "SPSSX(r) Discussion" <SPSSX-L@LISTSERV.UGA.EDU>
From: Hideko Bassett <hhamada@gmu.edu>
Subject: Comparing 2-factor and 3-factor models
In-Reply-To: <7.0.1.0.2.20080612123640.03b0d668@mindspring.com>
Content-type: text/plain; charset=US-ASCII; format=flowed; delsp=yes
Hi,
I'm using AMOS to compare fitness of 3-factor and 2-factor models. Our
theoretically based model consists of 3 factors. By holding one of the
covariances between 2 latent variables (setting it as "1"), I believe
I can create 2-factor model nested in 3-factor model. Then, there 2
reference variables (set regression weight as "1") in the combined
latent variables (one reference variable per latent variable). My
question is that when I hold one of the covariances, do I have to
change regression weight of one reference variables? I'm wondering if
by holding the covariance between 2 latent variables is the same as
one latent variable, then, there shouldn't be 2 reference variables in
it. So I tried to run analysis on this 2-factor model in two different
way. One is to leave 2 reference variables as they are, and the other
is to release one of the reference variables. The two results I got
were dramatically different (the first one had chi-square = 85 and
significant, and the latter one had chi-square = 21 and not
significant).
As you can see, I'm vary new to the world of SEM and AMOS, so any help
would be appreciated.
Also, our data is not normally distributed, so I would like to get
robust chi-square, but I'm not sure how to do it with AMOS. Does
anyone have suggestions on which estimation method to use with non-
normally distributed data?
Thank you very much.
Hideko
=====================
To manage your subscription to SPSSX-L, send a message to
LISTSERV@LISTSERV.UGA.EDU (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD
| 
