anne sophie posted on Thursday, September 18, 2014 - 11:09 am
Iíve got a model with a nominal independant variable (X, 4 levels), a nominal mediator (M, 3 levels) and a binary dependant variable Y. Is there a way to estimate this model with direct and indirect effects?
If not, I might be able to change the definition of X and M to ordinal. Would this help?
Iíve tried several syntaxes, defining the variables various ways, different estimation techniques and MODEL options, and lím not convinced of the meaning of the results I get. Iíve searched the references and discussions but I havenít found the confirmation that iím doing it correctly.
There are ways to do this, but it is advanced; see the paper on our website:
Muthťn, B. & Asparouhov, T. (2014). Causal effects in mediation modeling: An introduction with applications to latent variables. Forthcoming in Structural Equation Modeling.
The issue is with the nominal mediator. If it is reasonable to redefine it as ordinal you are better off if you are not at an advance level yet. As usual, the nominal X variable should be turned into dummy variables.
Erin Taylor posted on Wednesday, December 12, 2018 - 8:13 am
Dear Dr. Muthen,
Following on the above posts, would it be incorrect to turn the nominal mediator variable into dummy variables, and then use these dummy variables as regular categorical variables (specifying them as such in the CATEGORICAL ARE statement) for the mediation analysis?
I don't think the indirect effect would be correctly expressed by that approach. A rigorous way to handle a nominal mediator is described in Chapter 8 of our RMA book for which we have examples on our website.