Setting up contrasts - is it better to include as much or as few contrasts as possible?

Hello! I am setting up my contrasts through the FSL GUI. Does anyone know whether it’s best to keep my contrasts as lean as possible when setting up my design in FSL? I am not sure if including more contrasts would carve into the variance (whether it is a pooled variance).
I’m wondering whether I should include as many contrasts as I think I might need in the future or only those that I’d need for my model now (which I would keep lean if they pooled together).

Any advice/suggestions are appreciated. Thank you!


In terms of contrasts, you can look at as many as you like, but the more you produce, the more likely you are to explore results and come up with some kind of false positive. If you preregister analyses and only analyze what you originally planned to analyze, then this shouldn’t be a problem. Otherwise, you can control for false positives across all tests you looked at (including exploratory).

What you may want to think about keeping lean, especially for task-based scans that tend so be shorter than resting state, is what goes into your first level models (e.g. denoising). Including a lot of model confounds (for example, 36 head motion parameters, a bunch of AROMA/aCompCor components) reduces your temporal degrees of freedom. If you have a long scan with a quick TR than this is less of an issue.


Thanks so much Steven!