Hello, I have a data set on twins, so that for each twins there are 23 items on a 5-point scale (i.e., quasi continuous). I want to perform a FMM, similar to Muthen, Asparouhov, and Rebollo (2006). Before the heritability analysis, I want to decide on the right number of classes and factors. For that I would like to ask a few questions:
1. Is example 7.27 in the version 7 manual is the right example to look at (except that I won't need to specify the items as categorical)?
2. Does this analysis supposed to be done for all twins simultaneously (i.e., should I use also type=complex and organize the data in a 'twin below twin' format)?
3. Is example 7.27 relevant also for the next step, in which I need to compare this model to a model of 2 or 3 factors within each class?
Thank you so much for the answers. Your support is really helpful.
Regarding (2)-from Muthen et al. (2006) I understand that the wide format is preferable and I will follow this recommendation. However, I got complicated syntax wise. Is there any example that could help me in that?
If not, I would like to make sure I understand correctly. Lets say I have c1 and f1 for twin1 and c2 and f2 for twin2.
A) to ensure equalities of measurement across twins, should I assign the same number in brackets to equivalent observed items on the latent classes? for example:
%c1#1% [u11] (1) . . . %c2#1% [u21] (1) . . .
B) to let c and f correlate freely across twins, should I just write "c1 with c2" and "f1 with f2" below MODEL:%OVERALL%
another preliminary step I am supposed to do before turning to the Factor Mixture Analysis, is to examine the number of factors in a CFA.
The problem is that CFA is theory driven, and while I have a theory regarding 3 factors and 4 factors, I don't have any idea how to divide my observed items between two factors. Also, I don't think that the 3 factors and 4 factors solutions that I have in mind are nested, which is another problem.
I guess that what I ask is- what models am I supposed to compare in the preliminary FA? All the papers I am relying on when studying this method decided to go with a one-factor solution, so I never got the chance to see the next step...