ROI anlyses: combine or separate?

Dear experts,

I have a somewhat naive question about choosing ROI in roi-based-decoding analyses. I have several different criteria for selecting ROI: 1) regions found from previous literature 2) searchlight 3) my own theory-based prediction. Shall I put these ROIs in one mask and do roi-analyses all at once or shall I do it separately? What difference would it make?


The problem of trying different ROIs and reporting the best working one is that your results biased: this is known as selective inference. Please do no dot that.
If the different regions definitions overlap substantially, I would report the accuracies for all of them or the average accuracy. At least, this number is not biased.
Taking the union of ROIs does not really make sense to me.

The ultimate decision is: how should the reader interpret the numbers ? would he/she be able to reproduce the result ?

Hmm. I was just thinking that if taking the union of ROIs does not make a difference, at least it is more time efficient than doing it separately (It might sound funny). If possible, could you please recommend some papers that have used similar method such that define roi both base on theory and searchlight, etc.? Thanks.