Yes. You can label the direct effect in the MODEL command and use MODEL CONSTRAINT to create the indirect effect. MODEL TEST can be used to test if the two parameters are significantly different from each other.
Is there a way to test if the indirect effect in a SEM with multiple mediators is a significant proportion of the direct effect? If so, is there sample code?
Our SEM examines how an intervention works through multiple mediators to predict a set of outcomes. We want to see if the indirect effect through two of the variables (e.g. Intervention --> Var 1--> Var 2--> Outcome) was a significant proportion of the main effect (intervention--> outcome).
You can express this proportion in Model Constraint using labels for parameters in the Model command. Because this proportion estimate may have a non-normal sampling distribution you may want to use non-symmetric bootstrap confidence intervals.
Thank you so much for your response. I have a quick follow-up question: We were hoping to divide the indirect effect by the direct effect of a simpler model with no mediation effects (e.g. the original intervention effect on the outcome). Is there a way to test that in MPLUS? The "direct effect" of the intervention on the outcome is much smaller when the mediation pathways are included, and I think the parameter of interest would be the proportion of the direct effect (with no mediators) accounted for by the indirect pathway.
Is there a way to incorporate an effect from a different model (e.g. from different syntax), by somehow saving the effect and then calling it back into the final model syntax?
We have two models - Model 1 is just a simple model of the intervention on the outcome variables. Model 2 incorporates mediators and thus indirect pathways.
We would like to test a ratio where the numerator is an effect drawn from Model 2 (an indirect effect) and the denominator is drawn from Model 1 (the direct effect of the intervention on the outcome). The main issue is that the corresponding "direct effect" from Model 2 incorporates mediator variables and is therefore smaller than the "pure" direct effect of the intervention on the outcome, which is why we would like to use the "pure" direct effect from Model 1 in our ratio.
Is there a way to save the direct effect from Model 1 to be used in the Model 2 syntax? If so, is there code to look at for this?
Thank you so much for your continued input on this issue. I am worried that I have not been clear with my questions.
A reviewer has requested that we test what percentage of the direct effect is mediated by our various indirect effects.
Conceptually, is there a way to statistically test the proportion of a direct effect (from a model with no mediators) that is explained by an indirect effect. In our specific model we have an intervention, 4 mediators, and 3 outcomes. Is there any way to test if certain indirect pathways explain a significant proportion of the original direct effect?