Latent Class Mediation PreviousNext
Mplus Discussion > Categorical Data Modeling >
 Steven A. Miller posted on Tuesday, April 23, 2019 - 12:56 pm
I have a continuous outcome, a dichotomous exogenous variable, and a latent class that I theoretically believe to be a mediator.

1. Is it necessary to use causally defined mediation for this?

2. Should I perform a manual three step where I save class membership information and use it in the mediation with a known class?

3. If I do that, it seems that I can't have auxiliary outcomes as both predictor and outcome. Is it okay to use them as outcome in Step 1?

4. Does the MODEL INDIRECT work for this sort of model?

I have been looking at chapter 8 of your book, but it seems that the nominal latent class mediation is only with dichotomous outcomes.
 Bengt O. Muthen posted on Tuesday, April 23, 2019 - 5:03 pm
1. I think so.

2-3. I wouldn't go that way.

4. Yes, a nominal mediator can have any kind of outcome using IND.
 Steven A. Miller posted on Wednesday, April 24, 2019 - 8:00 am
Thank you! As you wouldn't go the manual three-step approach, Would you then do the estimation of all things in a single step? Is there an issue with class assignment because of covariates and outcomes? Or is there some other method?
 Bengt O. Muthen posted on Wednesday, April 24, 2019 - 4:55 pm
Q1: Yes.

Q2: Try a single-step approach and see if class formation shifts relative to the LCA-only part.
 Steven A. Miller posted on Saturday, April 27, 2019 - 9:12 pm
Should the MODEL INDIRECT output then list the class variable as mediator? That is, should there be a specific indirect effect that looks like Y C X in the output? My C is not listed and I'm wondering if I'm having an issue in my setup or if it doesn't get listed this way.
 Bengt O. Muthen posted on Sunday, April 28, 2019 - 1:05 pm
Model Indirect does not cover a nominal mediator. The Mplus approach to this uses Model Constraint and is described in Section 8.5 in our RMA book for which you see inputs for Table 8.35:
 Steven A. Miller posted on Sunday, April 28, 2019 - 1:52 pm
I see. And, since in this case the outcome is continuous, I could just alter the model constraints (and definition of [y$1], [y$2], etc.) to be consistent with those dealing with Table 17 in your "Applications of causally defined direct and indirect effects in mediation analysis using SEM in Mplus" paper? Do you have full citation information on this paper?
 Bengt O. Muthen posted on Sunday, April 28, 2019 - 3:53 pm
Right. That paper has not been published. See how it is referred to in
Back to top
Add Your Message Here
Username: Posting Information:
This is a private posting area. Only registered users and moderators may post messages here.
Options: Enable HTML code in message
Automatically activate URLs in message