Hi everyone,

I have a question regarding the correlation between predictors in a model, and whether I should worry about it in my specific case. Any help would be great. Here are the details:

I am interested in the effect of within-subjects treatment Z (two levels) on my dependent variable Y (which is measured under various conditions). I measured Y at the start of the experiment as a 'baseline' pre-treatment measure (Y1), and then measured again under each of the two levels of Z (so Y2 has two values per subject).

I am interested primarily in the effect of treatment Z on my DV, Y2. I want to take 'baseline' Y1 into account, so I can either A) use Y1 as a time-varying covariate in a mixed model with Y2 as the DV, or B) calculate Y2-Y1 and use the difference score as the DV. But I am also interested in how two other continuous variables, A and B, interact with Z to affect Y.

I have run both analyses, and I find a significant A * B * Z * Y1 interaction (in analysis A) and equivalently an A * B * Z interaction in analysis B. My problem is that A and B are related to Y1 (my 'baseline' measure of Y). There is a strong A * B effect on Y1. My question is, does this known A * B interaction on Y1 invalidate my analysis of Y2, in either or both options presented above?

Any advice would be highly appreciated, I am quite a beginner in my understanding of this type of statistics.