Lol thanks!!!

Also, I will be at #SRP this week if anyone wants to chat!

Current projects in the lab include longitudinal neuroimaging of substance use at different time-scales, family-based studies of casual and genetic effects, and the development of new ML models for task fMRI. This is a funded position with up to 3 years of funding available.

Thanks! I was able to create an educator account on datacamp, which lets me give trainees access for free if then enroll in my 'class'. So far it looks like a useful supplement, particularly for programming concepts that might be new

For sure. I guess my point is that a generative epistatic model with uncentered effects is equivalent to a centered model with large additive effects with the means added in after the data are generated. So the increasing additive effects you're seeing at higher MAF are expected.

Thanks so much!! Instead of saying that this is skew (MAF), I would say that this comes from not mean-centering X1/X2 before making the interaction term. I do find it surprising that the extent of the multi-collinearity (from not mean-centering) is larger when X1 & X2 are normal (MAF=0.5).

I'm finding it hard to wrap my head around this. I would have expected the opposite! Would you mind sharing your simulation code?

I posed the original as well! My re-statement is intended to explain why I'm interested in the question.

Ok thank you!!

I'm thinking of an additive effect as one in which the two variables are both significant predictors. It's not additive if one variable is unrelated, or doesn't explain unique variance. Model comparisons are also sensitive to when only one of the two is significant.