> Uncorrected at p<0.01 and corrected at p<0.05 is a possible
explanation.
> But
> in another map of this same paper, they also used a height and extent
> thresholds at p<0.001. In another paper, the author even used a joint
> height
> and extent threshold of p<0.0001. I think it's rare to use corrected p
> value
> at 0.0001, isn't it?
Well, let me step back and go over this again. To do a cluster-extent
correction, you somehow have to define clusters. In SPM, this is done by
applying a height threshold to a t-map. This creates a subset of
t-images with high-intensity voxels only. Clusters are then defined by
grouping these above-threshold voxels into spatially contiguous blobs.
So a height threshold is a necessary part of the cluster-extent
correction. Once the clusters are formed, then you can perform a
statistical test based on the extent of these clusters.
That said, I am not surprised you found people have used different
thresholds to define clusters: p<0.01, p<0.001, p<0.0001, etc,
(uncorrected I presume). This is because there is no "right" threshold
to define clusters. So people use whatever the threshold that suits
their data. Using lower (or higher) threshold makes your clusters larger
(or smaller) but does not make your clusters more (or less) significant.
That is because cluster-extent p-values are adjusted for the
cluster-defining threshold you used.
> Can anybody tell me how SPM calculates the corrected p-value?
You should probably start with
Worsley, K.J., Marrett, S., Neelin, P., and Evans, A.C. A
three-dimensional statistical analysis for CBF activation studies in
human brain. Journal of Cerebral Blood Flow and Metabolism,
12:900-918, 1992
K.J. Friston, K.J. Worsley, R.S.J. Frackowiak, J.C. Mazziotta, and A.C.
Evans. Assessing the Significance of Focal Activations Using their
Spatial Extent. Human Brain Mapping, 1:214-220, 1994
These are good intro as to how corrected p-values are calculated for the
height (Worsley et al) and cluster-extent (Friston et al). The current
implementation in SPM is based on more sophisticated results. See
Worsley, K.J., Marrett, S., Neelin, P., Vandal, A.C., Friston, K.J., and
Evans, A.C. A unified statistical approach for determining significant
signals in images of cerebral activation. Human Brain Mapping, 4:58-73,
1996
Perhaps you were thinking about calculating corrected p-values on your
own, but this is not a simple task to do.
-Satoru
|