There are a couple of points remaining here:
1. weights for F-contrast (unlike T-contrast) are defined as a matrix with
as many rows as effects of interest (eof): ones attributed (you compare
everything with everything else in both direction, so all significant
variance related to these regressors is extracted.
2. what is the point of defining constant term as your eof, it's just a
scalar: mean signal for the whole session, so there is no variance there at
all?
Why don't you want to get ts with variance related somehow to your task(s)
and regress out effects of no interest?
Iwo
On Aug 27 2011, Steve Fleming wrote:
>Hi Peter,
>
>Yes I think for your purposes it would be the latter (zeros(1:11) 1).
>
> As Iwo pointed out, the "adjust data" option removes the null space of
> your F-contrast, i.e. the zeros, from your timeseries. See lines 143-153
> in spm_regions.
>
>Hope that helps
>
>Steve
>
>On 27 Aug 2011, at 19:11, Peter Michalsky wrote:
>
>> Hello Iwo,
>>
>> first let me thank you for your reply. You understood correctly that I
>> want to regress out all nuissance factors while extracting a time-series
>> from a VOI. However, I also want to regress out the task variance (first
>> two regressors in the design matrix). Hence, I want to regress out all
>> 11 regressors and only keep the mean time-series. Do I do that by
>> ones(1:11) zeros(12) or by zeros(1:11) ones(12). If I understood you
>> correctly the correct solution should be to define an F-contrast with
>> all zeros except for the constant (zeros(1:11) 1). Maybe you could
>> explain to me why the opposite (ones(1:11) 0), which I thought would be
>> right is wrong. Thanks again for your help.
>>
>> Peter
>
>____________________________________
>Stephen M. Fleming PhD
>Wellcome Trust Postdoctoral Fellow
>http://web.me.com/stephen_fleming/web/Welcome.html
>
|