As for your question about randomise - you are right in that the gaps are where 1-p=0 (i.e. p=1). It is unusual to see values this extreme, but it is possible. The only interesting results from randomise are where (1-p)>0.95 in the corrected p-value output anyway. Any other values are not statistically significant.

On 22 Oct 2012, at 00:36, Jiyang Jiang <[log in to unmask]> wrote:

Dear FSLers,

I am re-posting this, since the previous post was somehow missed I guess. I need some suggestions on randomise output.

1). I use *randomise* to do a paired t test of hippocampal volumes at two time points, to see on which subfields the atrophy occured in the follow-up period. I tried both TFCE and voxel-based thresholding at both 1000 and 5000 permutations. I displayed the output in FSLview with intensity being 0~1, aiming to see the statistical outcome of the whole hippocampi. However, the boundary was not closing with some gap near the tail. My understanding of this gap was that the p value for these regions were 1, making (1-p) being 0, and therefore displayed as transparence. Was my understanding correct? Is there anyway to prove my thought? (P.S. the mask is closed)

2). I wanted to show the results in a 3D image. However, since the statistics were only showed on the hippocampal boundary, the display was hollow. And since the gap mentioned in the first question, there were a few leaks on the surface, making it looking ugly. I was thinking of projecting the result onto a hippocampi template. But it seemed that most templates would submerge the statistical significance. What would be your suggestion of the best way to present it in a journal paper?

I really appreciate any of your help!

Thanks and best regards,
Jiyang