testing for equivalence of latent mean structures in amos

#1
Dear all,

I´m using AMOS to anlyze mean differences between two different groups. I used (Byrne, 2010, Structural Equation Modeling with AMOS) as basic literature and followed the steps. I aded two groups and grouping variables, then I defined a model for both groups and used the multi-group-analysis button. This worked perfectly well. However, the second part (comparing groups) of the analysis did not work. To calculate the differences, one group has to be fixed as zero, whereas the other one is to be freely estimated. However, when I do the analysis, AMOS tells me "the model is unidentified. You will need to 5 add constraints". (5 is the number of latent variables)

I´m afraid, I don´t understand, why this happes. I really followed the steps, so I wondered, if there are any settings that have to be changes or anything I could do. i tried many times, but never made it work. Did anyone ever had the same problem? I would be very happy about your advice!

Thank you so much

Kathi
 

Kathi

New Member
#3
Have you imposed invariance across the groups on loadings and item intercepts?
Hi Lazar!

Thank you so much for your answer. I tried different ways. First, Amos automatically defines different models when using the multigroup-application and automatically equals loadings, intercepts, etc. Second, I did everything by hand and defined different model. I set the loadings equal, then the item intercepts , then the error terms, and finally the latent variable. So in the end I had 5 models from fully unconstrained to fully constrained as shown here (https://www.youtube.com/watch?v=H4gHIs4y6sU). However, unfortunately nothing worked out... I´m kind of clueless. Is there maybe any hidden property, I have to change beyond "estimate means and intercepts"?

Thank you for your help!

Kathi
 
#4
Hi Kathi,

I wonder if you ever figured this issue. I am conducting MGCFA and trying to test latent mean differences using AMOS but I am running into the same problem you described.

Please let me know whether you were able to figure this out.