AMOS:Modeling correlations between observed variables

[Tanja Gabriele Baudson]

Dear all,

I have two measurements (t1 and t2) of several variables (teacher
expectations, academic self-concept, motivation and GPA) over time and
would like to explore the relationships between them. The regression
paths from the t1 variables to the t2 variables are no problem;
however, I would like to model the correlations between the variables
within a point of measurement as well. The problem is that AMOS won't
let me -- it only allows correlations between errors, but not between
observed variables. I discussed this with a colleague (who said this
might be because the purpose of path models is to represent
regressions, such that AMOS is outsmarting the resesarcher) and had a
look at Byrne's book (who doesn't say anything on the issue) but find
neither satisfactory. As I think neglecting the correlations is wrong
from a theoretical stance, I would like to include them but have no
idea how. I'd be grateful for suggestions.

Thanks very much in advance
Tanya
--
Tanja Gabriele Baudson
Universität Trier
FB I Psychologie
Hochbegabtenforschung und -förderung
54286 Trier
Fon 0651/201-4558
Fax [phone]
Email [hidden email]
Web http://www.uni-trier.de/index.php?id=9492

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (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

[David Marso]

Hi Tanya,
  I haven't used Amos in a LONG TIME, but was using LISREL in the early 80's and Amos while I worked at SPSS.  Don't have any SEM package available at the moment.  One trick is to simply reparameterize the model so that you create a 1:1 relationship between the observed variables and the latents.
i.e Constrain error variances to 0.
Constrain the Factor loadings from Observed to Latents to Identity Matrix.
Model the Covariance structure of the Latent variables which is actually in this case equivalent to the observed.
Set up the same relation for the Y side and all the time based paths can be set up in the Gamma and Beta matrices.
X=I*Ksi + 0
Y=I*Eta + 0
Eta = Gamma*Ksi + Beta*Eta + resid.

HTH, David

[David Marso]

See attachedSEM.pdf