Message/Author 


I am an Mplus novice. I ran a latent profile analysis and identified 3 profiles. I then put profile assignment into a path analysis as an endogenous variables. I ran the path analysis with profile assignment as one variable and I also broke it into dummy variables. The model with dummy variables fit better but I'm having trouble interpreting the model results. My questions are: 1) Is it valid to turn a categorical variable into dummy variables? 2) How do I interpret the model results? For example, when I regressed my mediator (Eff) onto one of the dummy variables (Cls2D) I got a weight of .064. What does this mean? Is it like an odds ratio in logistic regression? 


Path analysis with a nominal DV is a tough task for any analyst, and certainly for a novice. Probably the easiest approach for you is to work with one 0/1 dummy variable at a time as the DV (one of the 3 categories vs the 2 other). But even so, you get into new, advanced methods because path analysis (by which I assume you mean mediation analysis) needs special attention with a binary DV using counterfactuallydefined indirect and direct effects (see our Mediation page on our website for sources to study). I assume that when you say "regress my mediator (eff) onto ...Cls2D" you really mean the opposite  that is, Cls2D is your DV which is regressed on the mediator (eff). 


Here is the syntax that I wrote: TchPrac on Eff; Eff on Cls2D; Eff on Cls3D; TchPrac = teacher practice Eff = Selfefficacy Cls2D = Class 2 dummycoded Cls3D = Class 3 dummycoded (I left out Cls1 to use as a reference) The main intention is to see how class membership predicts teaching practice, using efficacy as a mediator. (I had a few other mediators as well but the model fit was not good so I took them out.) Is there a better way to do this? 


These general analysis questions are better suited for SEMNET. 

Back to top 