AMOS: Troubles with single item latent variables

Hi all (this is my first post to the forum, so please be gentle... :))

I am working on a structural model in AMOS. The model includes two independent latent variables (both with multiple indicator items) and three dependent variables. The DVs have each been measured with a single item. At the moment, I am in the middle of validating the constructs (i.e. working on the measurement model; in other words, CFA).

I have created a latent variable for each DV ("pseudo latent") and as a part of this, I have specified the error term variances and factor loadings for them. Thus, instead of assuming the DVs being free of error, I use the latent factor procedure to allow the specification of measurement error.

However, I have a problem that I don't fully understand. If I specify the three DVs as directly measured variables in the model and thus free of error, then the correlations between the three DVs (as shown in AMOS output) are much weaker than if i used the latent factor procedure. Can anyone tell why is this?

Second, what is your opinion, should the three DVs be included in the measurement model at all or should I include only the IVs?

Br, MrNorris


Phineas Packard
I am not sure I 100% get you but it is well-known that manifest variables are typically biased toward zero when compared to latent variables. If I have understood you correctly the correlation between x and y is most likely going to be smaller than L(x) and L(y) when L(x) and L(y) are defined as latents with \(\lambda\) loading and residual \(1-\lambda^2\). I might be able to help more if you give more information on the model you are specifying.
Lazar; you understood correctly. For further clarification, see the attachment. So, in Model 1, we have dv1, dv2 and dv3 as directly observed variables, whereas in Model 2 we have L(dv1), L(dv2) and L(dv3).