I am using the MLMED macro to test a 2-1-1 moderated mediation model where repeated measures (Level 1) are nested within individuals (Level 2). Theoretically, my moderator is on Level 1 as it changes with time, but I want to model it as a Level 2 moderator in order to be able to use MLMED (MLMED cannot handle level 1 moderators).
My understanding is that they did something similar in the following article:
https://doi.org/10.1002/cncr.33850
This is, the moderators are on Level 1 (ie., fatigue, depression, physical activity change with time), but MLMED should have treated them as being on Level 2.
How should I implement this in my analysis? Should I create aggregate scores for the moderator variables (the average for each participant across time points) before running the analysis or should I use participants' scores at all time points? Originally, I thought that it wouldn't make a difference on the grounds that MLMED was using the average score of each participant across time points (since it's a between group effect on level 2). However, I have tried out both ways with my data and I get different results.
Practically, in which of the following ways should my dataset be structured?
Participant# - Time - Moderator
1 - 1 – score of participant 1 at T1
1 - 2 – score of participant 1 at T2
1 - 3 – score of participant 1 at T3
2 - 1 – score of participant 2 at T1
2 - 2 – score of participant 2 at T2
2 - 3 – score of participant 2 at T3
OR
Participant# - Time - Moderator
1 - 1 – average score of participant 1 across T1-T3
1 - 2 – average score of participant 1 across T1-T3
1 - 3 – average score of participant 1 across T1-T3
2 - 1 – average score of participant 2 across T1-T3
2 - 2 – average score of participant 2 across T1-T3
2 - 3 – average score of participant 2 across T1-T3
I hope my question is clear enough. Your help is very welcome!
