Hi Chobok,
If you run a one-sample t-test at the second level, of a bunch of
first level A>control contrasts, then I think that you effectively
pair each subject's conditions, which seems right. If you instead did
an unpaired two-sample t-test of first level A contrasts and control
contrasts, then you are ignoring this pairing. If you did a paired
t-test at the second level, then I would expect that to match the
results of a one-sample test of the differences (the contrast images
are essentially differences, or more generally linear combinations, of
beta images).
Similarly, I think if you pair properly (with either method), then
(A>C)>(B>C) and A>B will be equivalent.
I hope that helps,
Ged.
On 16/10/2007, chobok kim <[log in to unmask]> wrote:
> Dear SPMers,
>
> A few days ago, I posted a question about contrasts, but I could not
> receive any answer. I'm sorry for reposting this.
>
> If I have 3 conditions including two experimental (let's say A and B)
> and a control conditions, the t-maps for all conditions would be
> calculated by using simple t-test at the individual analysis. And
> then, I can specify "A>control" and "B>control" at the group level.
> But, in different way, I can make "A>control" and "B>control" using
> two-sample t-test at the individual level then I can run simple t-test
> for "A>B" or "B>A" as well as "A>control" and "B>control".
> In other words, at the following table,
>
> the first way the second way
> 1st level A, B & control (simple t-test) [A>control] & [B>control]
> 2nd level A>control & B>control [(A>control)>0],
> [(B>control)>0], [A>B] & [B>A]
>
> [A>control] in the first way and [(A>control)>0] are same?
> Otherwise, which way is a better way?
>
> I appreciate your help in advance.
>
>
> --
> Chobok Kim
> e-mail: [log in to unmask]
>
|