Wow - i deeply admire you for this decision and the sacrifices you are making! You just made my day.

Hi Adrien, it may be a while until we have something online…. But i would be happy to have a zoom call and walk you through what we are thinking right now.

So, please don't cite the original paper (
www.sciencedirect.com/science/arti...) without having read ours!

The main lesson: It is about 3x harder to publish a careful refutation of an influential idea, than it was to publish that idea in the first place. The paper as 5 experiments and nearly 700 subjects - plus an number of additional control experiments.

The main lesson is that "unique variance" is highly dependent on which other regressors we consider. That's a problem, as we never can be quite sure what the correct model is. Bayesian model averaging can at least integrate out some of that uncertainty. We hope to have the manuscript out soon!

I guess you could think about the "shared variance" as a lower bound for the overlap between regressors. However, for this to be a real bound, you would need a strong theoretical basis to distinguish overlap and suppression. There is none (I know of) - it just boils down to geometry (Fig. 3).