For my model, i'm using hayes process model 6 with 3 mediators. I hypothesize that my M1 mediates between X and M2/M3. To which effect should I look in order to answer this hypothesis? Or doesn't it make any sense to hypothesize that?
Should I transform my model for this hypothesis, into model 4? Where X will be X, M1 will be M1 but M2/M3 will be Y? or are there effects that I will be missing out for?
Thank you for answering in advance!
1) If you have a priori hypothesis that M1 mediates between X and M2/M3, then you should test exactly this model and not somthing else. If your hypothesis makes sense is up to you and your research. Nobody from outside without knowledge about your research can answer this question.
2) I am not sure about your model. You said " M1 mediates between X and M2/M3". Formulated that way, it does not sound like model 6 from Hayes but more like model 81 or something similar. Can you draw a path diagram so that we can see what your hypothesis is
3) Not all possible models are readily modeled with numbers in PROCESS, but it is possible to handle individual models via syntax command. With the "bmatrix" command you are able to specify which variable should mediate at which position. With the "wmatrix" and the "zmatrix" you can add moderators to the model. So in principal, all combinations possible (under the restrictions that has PROCESS, e.g. not more than 2 moderators).
Your conceptual model and your statistical model do not match. In your conceptual model, SMM has only a path to TW, but in the statistical model, SMM has a paths to TW, ITC and JS. This is not what you want.
Either you stick with the PROCESS macro and customize the model with syntax, according to your conceptual model OR you use structural equation modeling / path analysis. The advantage of SEM for such models is that you model all relationships at once and not via single regression analyses. This may be especially important, since you assume that some paths do not exist, i.e. are 0 (e.g. SMM->ITC and SMM-> JS). In SEM you have an implied covariance matrix, which will be compared to the estimated covariance matrix. If there would be a relationship where none should be, you would see this in the overall model fit.
So, for more complex models, like yours, I would prefer SEM, since it gives you more flexibility and more information about the (mis-)fit.
- Badges
- Science method
- Similar topics
- Mathematics
- Statistics
- Statistical Techniques
- Structural Equation Modeling