Print

Print


Hi Torsten,

Yes, this scales the values to a more interpretable level.  Say your study
had very long trials, then your BOLD activation would be higher than
another person who used the same stimulus with a shorter trial duration.
The scaling puts it into a framework where another investigator could
interpret it.  It is sort of like if I measured my office and said it was
3x3.  If you were an American you might guess my office is 3yards x 3 yards
whereas other may guess 3m x 3m.  This just takes out the guesswork
(because you'll report what trial type you used for the scaling).

If you had simple contrasts such that the sum of the positive parts (when
present) was 1 and the sum of the negative parts (when present) was -1,
then you do not need to do any scaling for the contrasts.  For example, if
you had c=(1 1, -1, -1), then you'd need to scale by 2, since the contrast
as is reflect the difference in sums instead of the difference in means.
This doesn't matter when we look at our test statistics, but when we
separate things out and look at the effect size alone it does matter.  If
you had c=(1, -1, 0, 0) then you're fine.

Hope that helps,
Jeanette

On Mon, Feb 6, 2012 at 9:48 AM, Torsten Ruest <[log in to unmask]> wrote:

> Thanks Jeanette! I hoped to just use the 2nd level copes (combined runs
> using fixed effects) to have them scaled to the same level, but from your
> message it seems I still need to include the heights from the 1st level
> runs...
>
> Thanks,
>
> Torsten
>