help
help > RE: FDR correction
Nov 30, 2015 04:11 PM | Alfonso Nieto-Castanon - Boston University
RE: FDR correction
Dear Alice,
Sorry the results table is not very clear, there are two levels of FDR-correction that you can use on your ROI-to-ROI results when selecting multiple seeds:"FDR seed-level" and "FDR analysis-level". The table shows you by default ony the "FDR seed-level" values, while you are using the "FDR analysis-level" values for thresholding (these are only shown in the results table when you select "FDR analysis-level" as a thresholding criterion as well). If you switch your thresholding option to "FDR seed-level" you should see there the same connections that were showing below .05 values in the results table.
To clarify, the "FDR seed-level" correction applies FDR separately for each seed ROI (it corrects across the multiple comparisons arising from having multiple target ROIs). This is akin to the level of correction that you would get in seed-to-voxel analyses (for a fixed seed ROI, correcting across multipel target areas). The "FDR analysis-level" correction applies FDR across the entire analysis (it corrects across the multiple comparisons arising from having multiple seed and target ROIs). This is typically a very conservative correction and only extremely strong results will survive. When looking at the entire matrix of ROI-to-ROI connections I typically suggest to try Network Based Statistics (NBS) approaches. For that, simply select an uncorrected connection-level threhsold (e.g. p<.001), enable permutation statistics and select one of the "NBS network" thresholds. That will evaluate whether the size of the suprathreshold subgraph(s) is significantly above chance (this is simmilar to the cluster-level statistics approach in seed-to-voxel analyses, where you fix your voxel-level threshold and evaluate whether the size of the resulting suprathreshold clusters is significantly above chance).
Hope this helps
Alfonso
Originally posted by Alice Yo:
Sorry the results table is not very clear, there are two levels of FDR-correction that you can use on your ROI-to-ROI results when selecting multiple seeds:"FDR seed-level" and "FDR analysis-level". The table shows you by default ony the "FDR seed-level" values, while you are using the "FDR analysis-level" values for thresholding (these are only shown in the results table when you select "FDR analysis-level" as a thresholding criterion as well). If you switch your thresholding option to "FDR seed-level" you should see there the same connections that were showing below .05 values in the results table.
To clarify, the "FDR seed-level" correction applies FDR separately for each seed ROI (it corrects across the multiple comparisons arising from having multiple target ROIs). This is akin to the level of correction that you would get in seed-to-voxel analyses (for a fixed seed ROI, correcting across multipel target areas). The "FDR analysis-level" correction applies FDR across the entire analysis (it corrects across the multiple comparisons arising from having multiple seed and target ROIs). This is typically a very conservative correction and only extremely strong results will survive. When looking at the entire matrix of ROI-to-ROI connections I typically suggest to try Network Based Statistics (NBS) approaches. For that, simply select an uncorrected connection-level threhsold (e.g. p<.001), enable permutation statistics and select one of the "NBS network" thresholds. That will evaluate whether the size of the suprathreshold subgraph(s) is significantly above chance (this is simmilar to the cluster-level statistics approach in seed-to-voxel analyses, where you fix your voxel-level threshold and evaluate whether the size of the resulting suprathreshold clusters is significantly above chance).
Hope this helps
Alfonso
Originally posted by Alice Yo:
Dear Alfonso
I encountered several times this case that I don't understand:
When conducting 2-level ROI-ROI analysis, in the results explorer I choose all of the seed ROIs and a 0.05 p-FDR analysis-level-correction.
I get no significant results.
Now, I choose instead 0.005 or 0.0005 p-uncorrected threshold and I get some results.
When looking at the results table, there is a column of p-FDR, and according to it there are results below 0.05 p-FDR....
I would very be grateful if you can help in clarifying this.
Many thanks!
Alice
I encountered several times this case that I don't understand:
When conducting 2-level ROI-ROI analysis, in the results explorer I choose all of the seed ROIs and a 0.05 p-FDR analysis-level-correction.
I get no significant results.
Now, I choose instead 0.005 or 0.0005 p-uncorrected threshold and I get some results.
When looking at the results table, there is a column of p-FDR, and according to it there are results below 0.05 p-FDR....
I would very be grateful if you can help in clarifying this.
Many thanks!
Alice
Threaded View
| Title | Author | Date |
|---|---|---|
| Alice Yo | Nov 30, 2015 | |
| Alfonso Nieto-Castanon | Nov 30, 2015 | |
| Abdullah BAS | Jul 19, 2019 | |