Hi Kosha,
You can deal with your outlier by using an EV that is one at the 154th
time point and zero everywhere else, regardless of what the rest of
the design looks like. It is always valid to do this, although you
should
make sure that convolution is turned off (as this is not a haemodynamic
effect) but that temporal filtering is turned on for this EV (as the
data
is temporally filtered, so the design needs to be too). As long as you
do this then it should all be fine.
As for the linear contrast - the [-3 -1 1 3] is exactly right. If you
did
[-1 0 1 2] instead then, as this contains a non-zero mean, it would
model a mixture of the mean effect and the linear effect which you
could not disentangle. By using [-3 -1 1 3], which has zero mean,
you effectively decouple these effects which is what you want.
All the best,
Mark
On 17 Nov 2008, at 02:00, Kosha Ruparel wrote:
> Dear Fslers,
>
> I have two seperate questions regarding design and data.
>
> Q1) dealing with outlier data points
>
> For a block design with 200 time points having a huge relative
> motion jump at 154th time point, is it valid to interpolate that
> time point with the 153rd and 155th time point and then process the
> data? i think in an event related model, one could add an additional
> EV with a 1 at the 154th time point. this would however not be
> possible i think for a block design.
>
> Q2) i have read the posts discussing the classic 3 condition
> parametric modulation case where the correct contrast is [-1 0 1].
> for a 4 condition parametric modulation, the contrast for a linear
> effect from A-> D would be [-3 -1 1 3 ]. is this correct? if one
> would model [-1 0 1 2], would that also be similar?
>
> thanks for the help in advance
> Kosha
>
|