I believe that your reading of the output is correct---with an uncorrected height threshold of p = 0.05, the probability of finding a cluster of size 1060 by chance is less than 5%. However, a cluster of 1060 3x3x3mm voxels is quite big, and does not afford enough resolution to identify different brain regions. A more reasonable approach would be to use a lower uncorrected threshold (e.g, p < 0.001), which would let you identify clusters with a smaller size at the same corrected alpha of 0.05. Note that the corrected p (i.e. alpha) value is always determined by two inputs, the uncorrected voxel-wise p-value and the cluster size. If you have direct access to AFNI, you can use the program 3dClustSim, which offers more options to identify a more reasonable pair of cluster sizes and uncorrectd p-values.

I believe that your reading of the output is correct---with an uncorrected height threshold of p = 0.05, the probability of finding a cluster of size 1060 by chance is less than 5%. However, a cluster of 1060 3x3x3mm voxels is quite big, and does not afford enough resolution to identify different brain regions. A more reasonable approach would be to use a lower uncorrected threshold (e.g, p < 0.001), which would let you identify clusters with a smaller size at the same corrected alpha of 0.05. Note that the corrected p (i.e. alpha) value is always determined by two inputs, the uncorrected voxel-wise p-value and the cluster size. If you have direct access to AFNI, you can use the program 3dClustSim, which offers more options to identify a more reasonable pair of cluster sizes and uncorrectd p-values.

Dear Andrea,

Yes, you're absolutely right. The clusters were really huge. I've just rerun the analysis for individual p-value threshold of 0.005. It looks ok now.

Thank you very much for this!

Just to add a bit with Andrea, let's say, you have 8196 voxels. If you set a whole brain/ voxel-wide P-value of 0.05 (which is fair enough), then it turns into the voxel-wise p-value as p

Alphasim is an afni program (updated version clustersim) in which you input a corrected p value and the output is the number of voxels in the clusters which surpass that value. I would not recommend analyzing your dataset across programs--spm and afni. Wonky things start happening!

Dear Christine,

Alphasim is a handy tool to run Monte Carlo Simulations. As long as you have smoothness estimated from your data and a corresponding mask you can surely run it cross-platform.

Hi,

What does frequency mean there ? Can you define all of the columns ?

Thanks