Hi Jim,

> So I take it a user can set 3dTtest to one sample
> (i.e. NO group comparison) simply for the purposes
> of generating a random-effect map of a single group?

Yes, your 3dttest script looks good.

> Also, your response to my second question
> (whether I could fairly use the -mean from 3dttest
> or 3danova to make groupwise maps for direct
> visual comparison and separate tables of
> activations) implies that all else being equal, the
> patient group with the greater number of subjects
> will have less variance with a higher n, and thus,
> the two -mean group maps would not be an
> apples-to-apples comparision. Is this your
> implication? That it is really necessary to recruit
> equal numbers of subjects in the two groups?

If you only have two groups, the mean difference test with -diff in 3dANOVA is the same as two-sample t test with 3dttest no matter the groups have the same or different number of subjects.

The group mean test with -mean in 3dANOVA might be slightly different from what you would get with one-sample t test with 3dttest, but most of the time the difference should be negligible. The subtle difference is due to the fact that pooled variance of the two groups is employed in 3dANOVA for group mean tests while one-sample t test focuses only on the data of single group. Although using 3dANOVA for group mean tests is generally valid, it is usually recommended to run one-sample t test for group mean test to avoid the possibility of variance hereogeneity across groups.

We will modify 3dANOVA soon so that 3dANOVA will become a real generalized version of 3dttest; in other words, the above subtle difference will be gone in the new version of 3dANOVA. That is, group mean tests from 3dANOVA will be the same as one-sample t test with 3dttest.

Gang