Laurens De Gauquier Using Split File function and Organize Output by Groups, you can get results for each group separately in SPSS. However, this doesn't work with Process Macro. You can do Multi-Group Analysis in Smart-PLS to run your serial mediation model and get coefficients for each group separately. Moreover, It will also provide significance for path variance/invariance.

When you have a nominal X variable with 3 categories, there will only be two regression equations/two slope coefficients. A third equation would be redundant as the regression coefficients reflect mean differences between the 3 groups. With three groups, two equations are sufficient to capture the group differences (overall effect of X) entirely, although different coding schemes will lead to different meanings/interpretations of the regression coefficients (e.g., differences from reference group vs. differences from grand mean).

Dear Laurens De Gauquier

These tutorials might help, have a look:

1. https://www.youtube.com/watch?v=r2_zw4G55X0

2. https://www.youtube.com/watch?v=ByuUyLtoTt8

Best Regards

Dr. Fatemeh Khozaei

thank you Christian Geiser for your clear answer. I am however in doubt about the fact that I cannot really tell (let's say treatment A is the control) how treatment B differs from treatment C? Does this imply that I would have to perform another model where I change the order of the treatments? (so that, for example, treatment B is the first treatment and the model makes differences of B - A and B - C?). I believe that estimating 2 times the same model is not the way to go? Or can I, and merge the findings altogether? Thanks for the help

You should pick the coding scheme that makes the most sense for your application/hypotheses. For example, when you use dummy coding, the most meaningful reference category may be the control group (if you have one). If not, then I agree that running multiple analyses with different reference groups is not necessarily a problem. The overall model fit and R-squared will be the same. The only thing to keep in mind is that when you run all possible group comparisons, you may have to control/correct for Type-1 error Inflation (multiple testing, alpha error cummulation) in much the same way as in ANOVA post hoc pairwise comparisons.