Maria Rueda posted on Wednesday, November 28, 2012 - 9:42 pm
Hi, I need to estimate a CFA model of the following form I = a +bL +cX + e. or in a general form to not assume linearity . I=f(L,X) I correspond to the observed categorical items (28 items in my data), L to the latent variable that I need a proxy for and X are other covariates that I want to condition for.
So in mplus, how I add those X variables? Would it be like: L by I_1.... I-28 I_1 on X I_2 on X . . . I_28 on X
? Thanks for your help. Those X could be thought as the conditioning variables in a plausible values approach.
Yes on your question. The c's are called direct effects in a MIMIC modeling context. But you cannot identify all the c's in addition to the b's. At least one c must be fixed at zero for each L. In NAEP, conditioning on x for plausible values has all c's = 0. BSEM can be used if you insist on free c's; see
Muthén, B. & Asparouhov, T. (2012). Bayesian SEM: A more flexible representation of substantive theory. Psychological Methods, 17, 313-335.
Maria Rueda posted on Thursday, November 29, 2012 - 10:52 am
Thanks a lot for your answer. What I really need is an estimate of the latent factor. Some sort of factor score. Is there any advice for the type of estimator: ML or weighted least square? Thanks
I always hesitate to recommend using factor scores, since SEM was invented to avoid such use. But if you want it, I would use Bayesian factor scores - see the plausible value literature. See our plausible value paper on our site:
Asparouhov, T. & Muthén, B. (2010). Plausible values for latent variables using Mplus. Technical Report.
Maria Rueda posted on Thursday, November 29, 2012 - 12:02 pm
Thanks again for your nice reply. Sorry to bother with one more question. For Bayesian factor scores, do I need mplus 7? And how can I find an example of the syntax? in the analysis should I put estimator: bayes