I have a single step multiple mediation model. In my analysis, the TE is 0.34, in which 0.22 is direct while 0.12 is total indirect effect (for 3 mediators all together).
is it safe to say that the exposure has more effect on the outcome without involving mediators?
Even though the mediators play a significant role but still they don't have as much effect as the exposure on the final outcome.
Hi Hanish,
Normally in Mediation, the effect X on Y is reduced towards zero upon the addition on M, being complete mediation if X no longer predicts Y when M is in the equation. In your case, your TE increases. I think you have a suppressive effect according to:
Conger, A. J. (1974). A revised definition for suppressor variables: A guide to their identification and interpretation. Educational and Psychological Measurement, 34, 35-46.
An option you could consider here is to calculate the proportion of mediation
using a 1 - c'/c , or c' - c (even more Robust according to Imai, Keele, and Tingly, 2010).
In any case, I'd suggest you to rethink your analytical approach.
One could use your claim above, but I think it is safer to go back to your theoretical framemork and consider the possibility of running a hierarchical multiple regression, instead. You can use as your first step your X and then add, using theoretical premises, your Ms in your current model, treating them now as predictors.
I hope this makes sense for you. If you want to keep your model, just be prudent by saying that your 3 mediators, altogether, have a significant effect, but slightly lower than your path C'. Your TE being increased with M suggests a suppresive effect... they are important, but less important than X. That's why I suggested to run a HMR.
Best,
-Pau
Pau Garcia Grau thank you for your input. that is an interesting article. But it is for multiple regression analysis, doesn't mention any application to mediation analysis.
But I got your point of suppressors variable, looks like a different view of labeling mediator variables.
Hi Hanish,
You should not only analyze your non-standardized β coefficients, but also the lower and the upper limit for each effect, assuming that If this confident interval crosses the zero value it is non significant value. This is even more important than the β value.
Supossing your direct and indirect effects are significant i would carry out a Sobel Test for checking wether the mediation model is a significant model. This test will help you to determine if the proposed model is weaker when introducing your mediation variable or not, so it could be a nice option to reply to your answer.
Kind regards.
- Badges
- Science topic