Hi there,
I am using nilearn.glm.threshold_stats_img to threshold my second level z-maps. I’ve tested it with different height_control (FPR and FDR) and different p-values.
I get no significant voxels when using FDR (p-values of 0.05 and 0.001) and I get some significant voxels when using FPR (p-values of 0.05 and 0.001). FDR thresholding with higher p-values gives similar results as FPR thresholding results with the p-values 0.05 and 0.001. However, I may be wrong but it seems wrong to choose a higher p-value (not “standard” like 0.05 or 0.001) that gives significant results with FDR correction ^^’
I understand that FDR correction is more standard (as more stringent than FPR) and more standardly used in reporting data (e.g. in papers), and that FPR is more lenient and not used for reporting data.
Q: However, I see that FPR is the default height control in the function, and I am curious about the usage of FPR correction, how would you see the use of FPR correction: when/why/in which cases would someone use it? I guess I am wondering what is the relevance of FPR correction then ?
Thank you a lot for your time ! Don’t hesitate if you need more details