help
help > RE: Additional correction to corrected p-values from gPPI analysis in order to adjust for having multiple seeds
Nov 25, 2021 01:11 AM | Alfonso Nieto-Castanon - Boston University
RE: Additional correction to corrected p-values from gPPI analysis in order to adjust for having multiple seeds
Dear Maite,
There are a few different ways to do this, for example (in increasing complexity):
1) bonferroni correction: you may run 9 separate analyses (one per seed) selecting a p<.05/9 cluster-size p-FWE correction in each individual analysis, in order to correct for having multiple analyses
2) single-analysis: you may select all 9 seeds simultaneously and select the 'any-effects' contrast across seeds/sources (i.e. eye(9)) in order to perform a single combined analysis looking at effects across any of the 9 seeds using any standard thresholding option (e.g. your original voxel-level p<.001 and cluster-level p-FWE<.05) without any additional correction (this analysis can then be followed by post-hoc analyses in order to investigate which of the seeds may be chiefly responsible for any found effects)
3) FDR correction: you may run 9 separate analyses (one per seed), in each analysis using only a heigh-threshold (e.g. p<.001 voxel threshold) and no cluster-level threshold, extracting for all the resulting clusters their cluster-size p-unc values (uncorrected p-values), concatenating those uncorrected p-values across all 9 analyses, and then applying a joint FDR-correction (e.g. using P=conn_fdr(p) with p the vector of all cluster-level uncorrected p-values) and then selecting only those clusters with FDR-corrected p-values below .05 (P<.05 from the formula above)
Hope this helps
Alfonso
Originally posted by Maite Aznarez-Sanado:
There are a few different ways to do this, for example (in increasing complexity):
1) bonferroni correction: you may run 9 separate analyses (one per seed) selecting a p<.05/9 cluster-size p-FWE correction in each individual analysis, in order to correct for having multiple analyses
2) single-analysis: you may select all 9 seeds simultaneously and select the 'any-effects' contrast across seeds/sources (i.e. eye(9)) in order to perform a single combined analysis looking at effects across any of the 9 seeds using any standard thresholding option (e.g. your original voxel-level p<.001 and cluster-level p-FWE<.05) without any additional correction (this analysis can then be followed by post-hoc analyses in order to investigate which of the seeds may be chiefly responsible for any found effects)
3) FDR correction: you may run 9 separate analyses (one per seed), in each analysis using only a heigh-threshold (e.g. p<.001 voxel threshold) and no cluster-level threshold, extracting for all the resulting clusters their cluster-size p-unc values (uncorrected p-values), concatenating those uncorrected p-values across all 9 analyses, and then applying a joint FDR-correction (e.g. using P=conn_fdr(p) with p the vector of all cluster-level uncorrected p-values) and then selecting only those clusters with FDR-corrected p-values below .05 (P<.05 from the formula above)
Hope this helps
Alfonso
Originally posted by Maite Aznarez-Sanado:
Dear Alfonso,
I am performing seed to voxel connectivity analyses and I am applying FWE cluster based correction procedures (voxel threshold p<0.001, uncorrected and cluster threshold p<0.05, cluster-size p-FWE corrected) for each connectivity analysis.
Since I have 9 different seeds, I would also like to apply FDR correction (or other correction) to corrected p-values from the gPPI analyses, in order to further adjust for having multiple seeds.
Is this possible to do this correction for multiple seeds within the CONN toolbox? If not, do you know what would be the appropiate steps to follow in order to correct p-values obtained from CONN toolbox?
Thanks a lot in advance.
Best regards,
Maite
I am performing seed to voxel connectivity analyses and I am applying FWE cluster based correction procedures (voxel threshold p<0.001, uncorrected and cluster threshold p<0.05, cluster-size p-FWE corrected) for each connectivity analysis.
Since I have 9 different seeds, I would also like to apply FDR correction (or other correction) to corrected p-values from the gPPI analyses, in order to further adjust for having multiple seeds.
Is this possible to do this correction for multiple seeds within the CONN toolbox? If not, do you know what would be the appropiate steps to follow in order to correct p-values obtained from CONN toolbox?
Thanks a lot in advance.
Best regards,
Maite
Threaded View
| Title | Author | Date |
|---|---|---|
| Maite Aznarez-Sanado | Nov 24, 2021 | |
| Alfonso Nieto-Castanon | Nov 25, 2021 | |
| Maite Aznarez-Sanado | Nov 25, 2021 | |