rick reynolds wrote:
> Perhaps another point to note is that the statistics output
> from 3dANOVA are based on the degrees of freedom. If you
> don't have very many subjects, it is likely all of your stats
> will be smaller.
>
> How many subjects do you have?
>
> While one might expect 5% of 'noise' voxels to exceed p=0.05,
> that assumes having enough subjects to make the distribution
> basically normal.
Thanks for the response, Rick. I have 12 subjects, but I am looking at the distributions across voxels. 60,000 voxels should be plenty to get an even distribution that would allow 5% of the voxels to exceed the p < .05 threshold by random chance, would it not?
Bob Cox wrote:
> You could alleviate this problem by stretching the CCs to be more normal. The simplest
> way to do this, and the traditional one (going back to the Babylonians) is to take the
> arc-tanh of each CC.
Bob, I tried this but it didn't change the answer significantly; the maps look identical with only marginal changes in the resulting t-statistics. A vast majority of my correlation coefficients are far from the 1/-1 boundaries, and the distribution looks (to my eye) to be roughly normal. Much like a distribution of beta weights. Under these circumstances, is assuming normality still a problem?
Thanks for the help,
-cdm