Nearly, but you should subtract the mean of EV2 so that it has zero
mean.
Steve.
On 8 Dec 2008, at 22:05, Lara Foland wrote:
> Thanks for your feedback.
>
> So to be clear, one way to look at the correlation of FA with fMRI
> is to put
> everyone in group 1. Can you confirm that the following setup is
> appropriate?
>
>
> Inputs Group EV1 EV2
> controlsubj1 1 1 .545
> controlsubj2 1 1 .463
> controlsubj3 1 1 .445
> patientsubj1 1 1 .335
> patientsubj2 1 1 .377
> patientsubj3 1 1 .399
>
> EV1 EV2
> allsubj, grp mean [1 0]
> poscorrel with FA [0 1]
> negcorrel with FA [0 -1]
>
> Again, thanks so much,
> Lara
>
> On Fri, 5 Dec 2008 07:39:29 +0000, Steve Smith <[log in to unmask]>
> wrote:
>
>> Hi - a couple of problems with this:
>>
>> - If you specify different group memberships for higher level FEAT
>> analysis, the different groups need to have "separable" designs, so
>> you would either need to put eveyone in group 1 (which doesn't
>> invalidate the model, just possibly makes estimation a little less
>> efficient), or split EV3 into two EVs, one for each group. In any
>> case:
>>
>> - COn3 and 4 are not correct - a basic test of correlation with FA,
>> using your current model, averaged across two groups, would be [0 0
>> 1]. However, for reasons above, AND in order to allow you to test for
>> _differences_ in the FA correlation between the groups, you should
>> split EV3 into two, one for each group, and then test for within
>> group
>> correlations with [0 0 1 0], [0 0 0 1], etc., and for differences
>> between the groups with [0 0 1 -1], etc.
>>
>> - You should subtract the mean of the FA values before putting in the
>> model. In the case of splitting into two EVs, you should demean each
>> group's FA values separately, and for each group's FA EV, the other
>> group's entries will be 0s.
>>
>> Cheers.
>>
>>
>> On 4 Dec 2008, at 23:51, Lara Foland wrote:
>>
>>> Dear FSLers,
>>>
>>> I would like some clarification regarding contrast setup for an
>>> unpaired t-test
>>> with additional EV for correlation.
>>>
>>> I am conducting a t-test on fmri images for subjects in two groups:
>>> patients
>>> and healthy. I am interested in the correlation of any fMRI
>>> differences with
>>> differences in FA values (taken from an ROI within a DTI scan for
>>> each
>>> subject).
>>>
>>> My design matrix is set up like the two sample t-test example in the
>>> FEAT
>>> manual page but with an additional EV containing the FA value for
>>> each
>>> subject.
>>>
>>> Inputs Group EV1 EV2 EV3
>>> controlsubj1 1 1 0 .545
>>> controlsubj2 1 1 0 .463
>>> controlsubj3 1 1 0 .445
>>> patientsubj1 2 0 1 .335
>>> patientsubj2 2 0 1 .377
>>> patientsubj3 2 0 1 .399
>>>
>>>
>>> EV1 EV2 EV3
>>> Con1 con>pts [ 1 -1 0 ]
>>> Con2 pts>con [-1 1 0 ]
>>> Con3 con>ptsmod [ 1 -1 1 ]
>>> Con4 pts>conmod [-1 1 -1 ]
>>>
>>>
>>> Con1 and Con2 are the contrasts to test for fMRI difference between
>>> groups. Con3 and Con4 are intended to test for fMRI change between
>>> groups
>>> that also correlate with a difference in FA. Are Con3 and Con4
>>> set up correctly?
>>>
>>> Many thanks in advance,
>>>
>>> Lara
>>>
>>
>>
>> --------------------------------------------------------------------------
> -
>> 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
>> --------------------------------------------------------------------------
> -
>
---------------------------------------------------------------------------
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
---------------------------------------------------------------------------
|