help
help > RE: false positives cluster level of inference
Jul 5, 2016 08:07 PM | Alfonso Nieto-Castanon - Boston University
RE: false positives cluster level of inference
Hi Mary,
Thanks for bringing this up. Yes, if you are using any of the latest releases of CONN (15h or above) non-parametric cluster-level statistics based on permutation/randomization analyses are readily available in the results explorer window (before that release non-parametric analyses were already available but only for surface-based analyses). Simply switch there (in the top-right corner) the option that reads 'parametric statistics' to 'non-parametric statistics' and that will run the corresponding permutation analyses and compute non-parametric cluster-size and cluster-mass statistics from those.
Just for reference, in the Setup.Options tab there is also a field to specify whether you want your second-level analyses to allow parametric, non-parametric, or both types of analyses (by default that field will be set to both). Last, if you have already run your second-level analyses using an older CONN release then simply re-run those second-level analyses with a newer CONN release in order for the 'non-parametric' analysis option to become active.
Hope this helps
Alfonso
Originally posted by Mary Newsome:
Thanks for bringing this up. Yes, if you are using any of the latest releases of CONN (15h or above) non-parametric cluster-level statistics based on permutation/randomization analyses are readily available in the results explorer window (before that release non-parametric analyses were already available but only for surface-based analyses). Simply switch there (in the top-right corner) the option that reads 'parametric statistics' to 'non-parametric statistics' and that will run the corresponding permutation analyses and compute non-parametric cluster-size and cluster-mass statistics from those.
Just for reference, in the Setup.Options tab there is also a field to specify whether you want your second-level analyses to allow parametric, non-parametric, or both types of analyses (by default that field will be set to both). Last, if you have already run your second-level analyses using an older CONN release then simply re-run those second-level analyses with a newer CONN release in order for the 'non-parametric' analysis option to become active.
Hope this helps
Alfonso
Originally posted by Mary Newsome:
Hi Alonso!
I was just wondering about the Eklund, Nichols & Knutsson (2016) article that suggests that the cluster level of inference in SPM doesn't protect against false positives, and suggests that results at the voxel-level are accurate, although conservative. The authors suggested a solution of a non-parametric permutation method, which is accurate regarding false positives and not overly conservative. I don't know about the technical feasibility of this, but I wondered if there could be some way to incorporate the SnPM toolbox with Conn. Another approach may be to only refer to the voxel-level results (and not cluster-level), but that may be too conservative.
Thanks very much for your thoughts, as always!
Mary
I was just wondering about the Eklund, Nichols & Knutsson (2016) article that suggests that the cluster level of inference in SPM doesn't protect against false positives, and suggests that results at the voxel-level are accurate, although conservative. The authors suggested a solution of a non-parametric permutation method, which is accurate regarding false positives and not overly conservative. I don't know about the technical feasibility of this, but I wondered if there could be some way to incorporate the SnPM toolbox with Conn. Another approach may be to only refer to the voxel-level results (and not cluster-level), but that may be too conservative.
Thanks very much for your thoughts, as always!
Mary
Threaded View
| Title | Author | Date |
|---|---|---|
| Mary Newsome | Jul 5, 2016 | |
| Jeff Browndyke | Jul 11, 2016 | |
| Alfonso Nieto-Castanon | Jul 13, 2016 | |
| David White | Jul 8, 2016 | |
| Roger Mateu | Jul 12, 2016 | |
| Stephen L. | Nov 23, 2017 | |
| Alfonso Nieto-Castanon | Jul 5, 2016 | |