Dear Rachel,
If you have a value (e.g. depression score) which is the same across
both of the scans that you are pairing in the t-test, then there is no
point including this score as the paired t-test is already controlling for
all possible changes between pairs (by removing the within-pair mean
and so only looking at the within-pair difference for the remaining stats).
If the value is different between the scans in the pair (i.e. there is a
within pair difference) then you could put all these values in a single EV,
but you cannot separate them into individual subjects (if this is what
the pairs are) since that would remove *all* signal from the input
(removing the mean and a difference within a pair leaves nothing else).
I hope this helps.
All the best,
Mark
On 15 Aug 2011, at 18:57, Rachel Kozink wrote:
> Hi Mark,
>
> Thanks for the response. As an extension of this problem (for which there were only 2 values for the regressor), how would one include a behavioral EV (i.e. depression score) into this design?
>
> Thanks
> Rachel
>
>
> -----Original Message-----
> From: FSL - FMRIB's Software Library [mailto:[log in to unmask]] On Behalf Of Mark Jenkinson
> Sent: Friday, August 12, 2011 8:10 AM
> To: [log in to unmask]
> Subject: Re: [FSL] dual regression within subjects
>
> Dear Rachel,
>
> Your last EV is redundant and should be removed.
> You can form this EV by a combination of the last 19
> subject-specific EVs. Hence it does not add anything
> and could be causing problems.
>
> All the best,
> Mark
>
>
> On 12 Aug 2011, at 12:57, Rachel Kozink wrote:
>
>> Hi Steve,
>>
>> That's correct - it's the model for randomise. I've attached the model. It's not clear from this image, but there are 38 subjects, each run twice so for the "group" variable, I have 38 different groups. As you can see, the last EV models which scanner the data was collected on and needs to go across groups.
>>
>> Thanks
>> Rachel
>>
>>
>> From: FSL - FMRIB's Software Library [mailto:[log in to unmask]] On Behalf Of Stephen Smith
>> Sent: Friday, August 12, 2011 2:33 AM
>> To: [log in to unmask]
>> Subject: Re: [FSL] dual regression within subjects
>>
>> Hi - I assume that you are talking about the model that you are feeding into randomise - probably best if you show us the model to comment on.
>> Cheers, Steve.
>>
>>
>> On 11 Aug 2011, at 14:24, Rachel Kozink wrote:
>>
>>
>> I am setting up a dual regression for a within subjects design and would like to add in an additional covariate. The scans are pooled from two different scanners (although each subject had both of their scans on the same scanner). When looking at the ICA components, a couple components have a marked difference within the group of subjects that correlates to which scanner the data was collected on. So, we would like to include scanner as a nuisance covariate in our paired t-test. When setting up my design, I receive an error message that states: "Warning - design matrix uses different groups (for different variances), but these do not contain "separable" EVs for the different groups (it is necessary that, for each EV, only one of the groups has non-zero values). " It seems to allow me to continue, but I wanted to verify that the results that I will get will be valid.
>>
>> Thanks in advance,
>> Rachel
>>
>>
>>
>> ---------------------------------------------------------------------------
>> Stephen M. Smith, Professor of Biomedical Engineering
>> Associate Director, Oxford University FMRIB Centre
>>
>> FMRIB, JR Hospital, Headington, Oxford OX3 9DU, UK
>> +44 (0) 1865 222726 (fax 222717)
>> [log in to unmask] http://www.fmrib.ox.ac.uk/~steve
>> ---------------------------------------------------------------------------
>>
>>
>>
>> <paired_design.png>
>
|