Need help with analyzing data from within subjects design study I am planning to conduct a within-subjects design study, and I am wondering if data analysis can be done using multi-level analysis.
IV and DV will be continuous variables. Moderator will be binary, dummy coded variable.
I have two levels – person level and condition level. I will measure the IV - pride once, so pride is a level 2 predictor.
I have 2 measurements of DV - altruistic behavior. This variable will be measured in two experimental conditions (norm violation vs control condition).
So my hypothesis is that the impact of pride on altruistic behavior is moderated with experimental condition. So that the influence of pride is stronger in the norm violation condition than in the control condition.
However, I am a bit confused, will the experimental condition be a level 1 moderator in this case? Can this analysis be done with multi-level analysis? Because if I remember correctly if you have a level 2 predictor, then you can not include a level 1 moderator in the model.
I am planning to do analysis in R studio with the lme4 package.
 A: I get terribly confused by the "level 1," "level 2" etc. terminology. If you are using lme4 you shouldn't have to worry about that.
You specify your fixed-effect predictors, you (ideally*) name identifiers for individuals and groups used for random effects so that there's no confusion about more than one indivdiual/group with the same identifier, and specify the structure of the random effects you want to include. The software will then provide the appropriate results.
A moderator is the same as an interaction term in a regression model. So you have a simple model with pride (continuous), condition (binary), and their interaction as fixed-effect predictors. With only 2 observations per individual you can include a random intercept for individuals. You evaluate the moderation via the coefficient for the interaction term.

*If there's a nesting structure like individuals within groups and you specify that structure explicitly to the model, the software should be able to distinguish individuals with the same identifier but in different groups. But why take a chance on getting that wrong?
