I have a model having 2 independents, 2 dependents, 2 mediators and 2 moderators. Which approach is scientifically more appropriate for hypotheses testing for such moderated mediation model? Kindly Reply.
Thanks in advance.
Dear Rehan Ahmad
I would suggest to have a look into Dr. James Gaskin tutorial on this topic and follow the same.
Kindly find the links below:
https://youtu.be/yMGkluhHxQY
https://youtu.be/glJhg_fxBag
Hello Rehan,
any regression based model (OLS type), may it be simple regression or mediation, is rather noninformative for the thing you want to know (causal direct and indirect effects). Hence, I would not recommend using Process (sorry, Cristian :). Better switch to a SEM framework that allows to a) test a restricted model which reduce the number of alternatives, b) and to incorporate instrumental variables for the IV and mediators (i.e., variables that have only effects on the mediator but not to the outcome). This may be difficult and often it fails but provides strong options to support your mediation assumption.
A OLS type mediation model is always saturated (has indirect and direct effects). Hence, a simple model with 3 variables (IV, mediator, DV) has over 40 (!) alternative structures which you cannot differentiate (see, Kline, 2015). Going to SEM allows testing restrictive versions with fixed-to-zero effects (e.g., full mediation or the mentioned instrumental variables).
If you allow me a shameless self-citation. We did that in an experiment (yes mediation is a problem even in randomized experiments!)
Jakob, E. A., Isidor, R., Steinmetz, H., Wehner, M. C., & Kabst, R. (2019). The other side of the same coin–how communal beliefs about entrepreneurship influence attitudes toward entrepreneurship. Journal of Vocational Behavior, 112, 431-445.
Kline, R. B. (2015). The mediation myth. Basic and Applied Social Psychology, 37(4), 202-213. doi:10.1080/01973533.2015.1049349
Best,
Holger
