I am new to MPlus and try to run a multilevel analysis including seperate survey items for different groups to model latent variables on three levels.
I would like to include data from student survey 1: x1-x10 (+ School ID & federal state ID) teacher survey 2: y1-y6 (+ School ID & federal state ID) school survey 3: z1-6 (+ School ID & federal state ID)
I would like to get estimates for the correlation/influence for model of: latent student performance by test items f1 by x1-x3, f2 by x4-x6, f3 by x7-x10
f1 on f2 f3 f2 with f3
while controlling for influence of teachers from the same school g by y1-y6 [all responses categorical] f1 f2 on g
and school factors (from survey 3 -> cluster type #1?) h by z1-z6 (all continious) f1 f2 on h g with h
within each federal state (-> cluster type #2?)
- How do I include responses to different items in seperate files?
- How do I have to phrase the WITHIN/BETWEEN to account for influences between the five latent variables spread over the three levels?
- How would I account for student gender in this model?
If only parts of this are possible to account for, which and how?
So you have 3-level data so you would use Type=Threelevel- or perhaps 4 levels if you count state. Mplus doesn't model 4 levels unless one is repeated measures but state can be handled by Type=Complex Threelevel if you have more than 20 states. The measures for students presumably vary across teachers and schools so they appear on all 3 levels. So you control for teacher variables on level 2 and school variables on level 3. See our Short Course Topic 10 on threelevel modeling.
You don't use separate files in Mplus - just one file with the x, y and z variables as columns (y and z variable values get repeated for lower-level units in the same cluster).