Sorry Mplus newbie here - is it possible to use latent variables in CFA as a predictor of a larger factor?
Yes you can, for example you have two factors (f1 f2), and want to add another factor as predictor (p1), each is measured with three items . Syntaxes might look like this:
CFA:
f1 BY item1-item3; !define factor 1
f2 BY item4-item6; !define factor 2
SEM:
f1 BY item1-item3; !define factor 1
f2 BY item4-item6; !define factor 2
p1 BY item7-item9; !define predictor 1
f1 f2 ON p1; !factor are regressed on the predictor
f1 WITH f2; !allowing the correlation between f1 and f2 after accounting for p1, mplus default
Hope that helps,
Manuel
I guess you are referring to the possibility of testing a higher-order factor model in which there is a so called formative structure between the first-order factors and the higher-order (general) factor, while the relationship between indicators and the first-order factors is reflective. If that is the case, no, you can't do it with Mplus. But it has nothing to do with the software itself. There is no direct way to do it within the traditional SEM framework although there are some workaround options. But I personally do not find them satisfactory. And, it is not a "true" factor analysis anymore, if you conceptualize your first-order factors as the predictors of your higher-order, general factor.
Ok so the issue that I have is directionality essentially I did a literature review to see what suggestions were made as to why Eating Disorders are higher in a specific population - I pulled out these suggestions and made them into a questionnaire with a 5 point Likert, turns out these questions cluster around 4 specific issues so now I want to be able to say these factors are predictors of Eating Disorder development
- Badges
- Science topic
- Similar topics
- Mathematical Sciences
- Statistics
- Statistical Software
- SPSS