[AMOS] - Model with many positive degrees of freedom not identified

#1
I am using a longitudinal structural equation model with several periods (side question: how can I enlarge the space to draw my models? max seems to be DIN A4...).
I have a latent construct that is measured by 3 indicators.
Let's look at the case with three periods:

- I specify covariances between the error variance from the indicators across each period over time, but only from the same indicator --> 9 covariances
- I specify covariance between the latent constructs --> 3 covariances
- I fix the mean of all the constructs to 0 and the variance to 1

This leads to a model with 54 distinct sample moments and 39 distinct parameters to be estimated, meaning 15 degrees of freedom.

Still, AMOS tells me, that the model is not identified and needs one additional constraint. I am not even sure that I should specify mean and variance for each period, I believe fixing it in one period should suffice...

Any help is greatly appreciated.
 
#3
Let's try with a path diagram then :):
For Brevity I included only periods 2006 - 2008, it should eventually go until 2011 (drawing covariances there is a mess...)

Also, thanks for the link, just fyi, I am using the fixed factor method (see Little, 2013, longitudinal structural equation modeling)
 

CB

Super Moderator
#4
It's hard for me to pick out a specific path that's causing the problem, but the really unusual thing about your model is the number of error covariances you've got going on. What's that about? Others may have more specific advice hopefully.
 
#5
The error covariances link the error of each indicator with its realizations in other periods. Always just specifically for each indicator. (Compare Little 2013, Longitudinal Structural Equation Modeling, page 162)
Do you know any other places where I might get help for this?