Hello,
>
> Now imagine I have 3 conditions and this 4th implicitly modelled condition
> If I wanted to say compare areas activated by condition 3 more than
> condition 1 and thus I DO want to ignore the implicitly modelled
condition,
> I'd enter contrast weights -1 0 1 (if I had specified 4 conditions I would
> have entered -1 0 1 0)
> Am I right so far?
Yes
>
> But what if I WAS actually interested in the hidden implicitly modelled
> condition?
> What if I wanted to know areas activated by condition 1 more than this
> hidden condition (and vice versa)
> In the above situation I am assuming the default for the hidden condition
is
> 0
> What would the appropriate contrast weights for these situations be?
>
Here, the contrast weight would be (1 0 0). Or, if you were interested in
where the hidden condition produces a relative activation in comparison to
condition 1, it would be (-1 0 0).
>Or in this scenario where you ARE interested in the hidden condition would
> you be better off specifying all conditions in the model?
>
Since the effect of any given condiition is meaningful only with respect to
the other conditions (i.e. it is expressed as an activation that is produced
by the comparison of one brain state with another) then you don't have to
model the last condition to represent its effect in an activation map. The 1
0 0 contrast implicitly compares condition 1 with the unmodelled condition.
One thing, though, that has exercised me a little concerns the ways in which
unmodelled conditions are treated. I assume that they are treated as flat
(base)lines. Does this mean that, if the unmodeled condition is actually
provoking a response that perhaps conforms to the canonical HRF, will this
non-flatness (excuse the term, not sure how else to refer to it) be
consigned to the residual error term (thus producing a structure in the
variance which, of course, we would like to avoid)? Any answers appreciated
Paul
|