Hi Hedvig,
> So in the difference analysis I assume it will assess
> differences in direction (e.g., activation in grp 1,
> deactivation in grp 2).
No, it seems that -expr 'step (a-a_threshold) - step (b-b_threshold)' would give you the following:
if the value at a voxel is 1, activation in group 1 but no activation in group 2;
if the value is -1, activation in group 2 but no activation in group 1;
if the value is 0, both groups are either active or inactive at the same time.
> If I want to assess differences also in magnitude (e.g.,
> strong activation in grp 2, weaker but significant
> activation in grp 2), does it make sense to remove the
> "step" function and only do for example
>
> -expr '(a-a_threshold) - (b-b_threshold)' ?
Well, it does give you sort of the significance difference between the two groups, but the problem is how to interpret the result: the difference between two t distributions is an unknown statistic.
Gang