1 sample t-test Tscore over 80 for specificity.

Submitted by vvillalta on
 Dear all, 
We are using DPARSFA in order to make an alff/falff analysis with a sample of 89 subjects. When we go for a second level analysis, FWE<0.05 results are a black mass of significant results all over the brain so we have to raise the T threshold up to a value of 80 in order to gain some specificity. Are we doing these analyses correctly? How do we set a t-value threshold that would give us significant information (not only significant results)?
Thanks very much in advance for any help.
Best,
V
 

YAN Chao-Gan

Tue, 04/03/2012 - 17:39

 When you perform ONE sample t-test on ALFF/fALFF, you will find p = 0.05 is not very good threshold, since most of the voxels contain signal and have ALFF bigger than zero (or global mean if mALFF used).

This also happens when you perform ONE sample t-test on functional connectivity maps. Most of the voxels will be significantly correlated with your seed.

Since ONE sample t-test can not tell you the difference between groups (you need TWO-sample t-test or ANOVA), then you can set a very high threshold (e.g., 0.001) for visualization.

Forums