have to calculate a moderated mediation with 2 mediators and 2 dependent variables.
My model:
predictor (dichotomous)
moderator (dichotomous)
both mediators (continous)
both dependent variables (continous)
My professor told my first to check my model with a MANOVA. After that I calculated my moderated mediation model with PROCESS (Hayes).
My problem is that I get different main effect results! Interaction significance are in MANOVA and PROCESS the same.
For example:
MANOVA: predictor is significant, moderator not
PROCESS: predictor is not significant, moderator is significant
But If I mean center in PROCESS my model I got similar results. But many people say to me that mean centering both dichotomous variable because it doesnt make any sense.
What is the problem?
Morning Christopher,
I think the problem is the different calculation procedure. If I understand your supervisor correctly, the MANOVA test was meant not as the main test, but just to check the effects direction. Moreover, MANOVA allows you test for simple moderating effects and provide additional statistics. If you have a complex model, however, (mediated moderation, moderated moderation, moderated mediation, etc.), then you need PROCESS.
So just stick to your results from PROCESS test. You can alternatively try the structural equation modelling, but I am not sure you need it.
Sincerely
Eugen
Centering the dichotomous variables (which is the correct way to go) is standard practice in the general linear model (anova, regression, manova). SPSS centers them when it computed the F-test in GLM e MANOVA. That is why you get the same results in process after you center yourself. Contrasts analysis, just to mention one, is based on centering categorical variables.
- Badges
- Science topic
- Similar topics
- Mathematical Sciences
- Statistics
- Statistical Software