Dear Torsten,
> thanks for this. In other words, in each image (tstat_1 and
> tstat_2) is the
> same information. The pos / neg eg in tstat_1 would reflect
> control>patient
> and control<patient respectively, I guess.
Yup.
>
> I've just another simple question, as you mentioned 1-tailed
> testing. I
> would like to express the stats as a function of t-value. Therefore
> I wanted
> to check the old fashioned tables for alpha=0.05. I have 6 subjects
> in each
> group (2 groups). Till yesterday I thought it would be a 2-tailed
> test, so
> my critical t value would be 2.228, but for a 1-tailed it would be
> 2.015.
>
> Could you just confirm 2.015 to be right?
It depends. If your "two groups" is indeed one group scanned under
two different conditions, i.e. if the # of subjects in your study is
6, then 2.015 would sound about right. If you have two "actual"
groups with a total of 12 subjects to which you fit two means then I
think it should be 1.81.
More important than the numbers though is if you should actually use
a one- or two-tailed test. The "philosophy" behind the one-tailed
tests in neuroimaging is that if you for example put in a contrast [1
-1 0 ...] to look for higher activity when seeing primed vs non-
primed words (just an example) you should not be "penalised" for the
possibility that you might later do another contrast [-1 1 0 ...] to
look for lower activity for primed words.
However, if you actually do that. I.e. if you test both the positive
and the negative contrast then you are in fact performing a two-
tailed test. And strictly speaking you should that that into account
(simply by entering 0.025 in your FSL check-box).
So in your case, if you really are interested in pat>ctrl and
pat<ctrl you should really use a two-tailed test.
Good luck Jesper
|