With this additional model comparison, the fit of the interaction is only marginally better than the model without. What I could find online was conflicting. According to some sources, the fact that the LOO difference is larger than 4 (see table above) is sufficient to argue that the interaction model is better. Other sources argue that the variance should be taken into account. In this specific case, we can see that the interaction term does not overlap 0, which is why I interpreted the results the way I did. But this is something I'd like to run by everyone in the dysco

I have an idea on how to improve Figure 3B. I think we can compute a decision boundary reflecting the interaction term, which should result in a diagonal line going from the top left to the bottom right reflecting the fit of the model and showing the separation between accept and reject off diagonal

I still need to find the appropriate reference here. I know where to look, just haven't gotten around to it just yet

I double checked in neuroimage. The article was published under creative commons license, meaning all figures can be reused given proper credit is given. We can discuss whether we want to change anything