Abby Gray posted on Tuesday, November 19, 2013 - 3:16 pm
I am new to both MPlus and SEM, and working on my dissertation. Please pardon this newbie question.
I am specifying a CFA model with 4 factors. In the structural model, the first 3 of these will be treated as exogenous and the 4th will be endogenous. The indicators for all of the factors were drawn from 4 different survey instruments--4 different individuals associated with each of 200 schools were surveyed about a certain issue within their schools. Items from 2-3 of the survey instruments are presumed to load onto each factor (for example, items from both the principal survey and the teacher survey are specified to load onto F1, as theory and prior research suggest that both of those observers experience F1 within their schools).
Because of the likelihood of common method variance, I plan to use a CTCU measurement model. This is straightforward enough. What seems a bit awkward is that the first 3 factors (which will ultimately be treated as exogenous) share common methods with the 4th factor (which will be treated as endogenous in the structural model). This makes for an ugly diagram, as there will be correlated errors between indicators on endogenous and exogenous variables. Should I be concerned about this? If so, how would you suggest that I address it?
Abby Gray posted on Tuesday, November 19, 2013 - 4:28 pm
Thank you very much. That is extremely helpful. Another question about this: I have a number of indicators for each of the 4 factors, and all of them will have errors correlated with several other indicators under the CTCU method. I would think that this jumble of correlated errors would make model identification almost impossible? Yet the literature suggests that CTCU is less susceptible to identification problems than other approaches to dealing with MTMM data. What are your thoughts on this?