I’m trying to simplify my data analysis, sorry this is a silly issue, but I just can’t quite decide what is the right thing to do.
My data is: ~400 participants, each providing ~150 behavioural measurements (within subject), as well as psychological questionnaire measures (2 per subject) What I’ve done previously:
a) regression predicting behaviour, like so: behaviour ~ 1+ reg1 + (1+reg1|subj)
b) show a moderation effect, i.e. that the impact of reg 1 on behaviour is moderated by the two questionnaire measures, like so: reg 1 ~ mo(psych 1) *psych 2 (no longer hierarchical as now I just have one row per subject; I’ve coded psych1 as ordered factor as there are only 6 possible answers)
I’ve been trying to simplify this to one single analysis, but I’m not quite sure which one of several is the right way to do it:
a) behaviour ~ 1+ reg1*mo(psych1)psych2 + (1+reg1|subj)
b) behaviour ~ 1+ reg1mo(psych1)psych2 + (1+reg1mo(psych1)*psych2|subj) - given that I only have one measurement of psych1/2 per subject, this is probably not right?
c) behaviour ~ 1+ reg1 + (1+reg1|subj) + (1+reg1|psych1:psych2) - given that there are many levels of psych1/2, this is probably not right?
Relatedly, for a) I now get a hard to interpret 3-way interaction term that I thought I could clarify using marginal plots like so:
But it tells me "Error: monotonic predictors must be integers or ordered factors. Error occurred for variable ‘Psych1’ "
I’d be very grateful for any advice.