Dear Richard,
> Dear Karl,
>
> I am confused by your reply to Luo. Doesn't a contrast of -1, +2,
> 0.... ask whether the parameter estimate for the second column is
> significantly greater than HALF the parameter estimate for the first
> column? Which is clearly different from the contrast -1, +1, 0 ...
>
>
> >> I would like to know how to define the weights of t-contrast
> >> in spm99. For example, what is the difference between c'= [-1, +1,
> >> 0,...] and c'= [-1, +2, 0,...] in spm99?
> >
> >They are the same. The contrast specification is exactly he same
> >in all versions of SPM.
I am sorry. What I meant to say was that a t-contrast in SPM99 is the
same as a t-contrast in SPM96 (I misread the question).
[-1, +1, 0,...] and c'= [-1, +2, 0,...] are not the same. The second
is equivalent to [-1, +1, 0,...] plus [0, +1, 0,...] (i.e. the
difference plus the second = weighted difference). This contrast is
testing a different hypothesis, one which is less easy to motivate
intuitively.
Sorry about that - Karl
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|