 |
 |
Hi, This is not intuitive but it is correct and does happen sometimes, just not that often. If all that was done was to apply the threshold then what you are expecting would be true - the clusters get bigger as the threshold is lowered (total number of voxels increases). However, there is also the calculation of a p-value for each cluster and this depends, in quite a complicated way, on both the size of the cluster formed and the threshold used to form it. Since we then threshold the cluster p-values themselves, rejecting anything above 0.05, then you can find that clusters can be significant at a higher threshold (because their p-values are better, due to the higher threshold) compared to a lower threshold, even though the cluster size is likely to be smaller at the higher threshold. Most often the cluster size shrinks as the threshold increases at such a rate that they balance or that the size change is more important and makes the p-values less significant. However, if the size change is small as the threshold increases then the effect on the p-value is likely to be that the p-value becomes more significant for the higher thresholds, and this might be enough to change it from being non-significant to being significant. I hope this helps. All the best, Mark On 20 Jan 2014, at 15:31, Dan Lurie <[log in to unmask]<mailto:[log in to unmask]>> wrote: I'm experiencing a weird issue with easythresh and was wondering if anyone else had seen it or can explain it. Specifically, for some statmaps, running easythresh at a higher Z-threshold (e.g. 2.1) results in *more* clusters than running at a lower threshold (e.g. 1.65). This does not happen for all maps, but is consistent (if it happens once on a map, I can reliably reproduce it by re-running easythresh at different thresholds). This doesn't quite make sense to me. As I understand it, lowering the threshold should result in more clusters, not less. Any insight is appreciated. Thanks! -- Dan Lurie Research Assistant Center for the Developing Brain Child Mind Institute http://www.childmind.org/