I'm not sure making the frequency dimension continuous will make
things any better as you'll have one continuous dimension and two
discrete and that's even less natural thing for SPM. Also the penalty
might be worse than Bonferroni. You can see an example for creating
nii files in spm_eeg_convert2images starting from line 169. I'm not
familiar with any special use of graph theory to correct for multiple
comparisons but you might inquire with people who work on connectomics
as they are the experts on that.
Best,
Vladimir
On Mon, Jun 25, 2012 at 12:54 PM, Elżbieta Olejarczyk
<[log in to unmask]> wrote:
> Dear Vladimir,
>
> The frequency dimension can be taken from 1 to 45 Hz, every 1 Hz. What is
> the procedure to convert the 18 x 18 x 45 data to .nii images ?
> What do you think about using of the graph theory to perform the statistical
> analysis of such data?
>
> Best,
> Elzbieta
>
>
>>Dear Elżbieta,
>>
>>One of the main reasons to use SPM for some data is that one can do
>>correction for multiple comparisons using random field theory.
>>However, that only makes sense if your data can be seen as samples
>>from a continuous random field. In what you describe there is no
>>continuity in any of the dimensions. The frequency dimension could
>>potentially be continuous but you discretised it into 5 bands. So you
>>can account for multiple comparisons in your data by either doing
>>bonferroni correction or using voxel-wise FDR. This could be done in
>>SPM but I'm not really sure SPM is the best software in this case as
>>it was designed for images and your data is not an image. You would
>>have to write some custom code to make sense of the results. In any
>>case you would either have to use a function that takes the data as
>>input such as spm_ancova or you should save your data as .nii images,
>>probably 18 x 18 x 5 and then you'll have 14 images per subject. But
>>again the way SPM presents the results wouldn't make sense for these
>>data as it assumes continuity.
>>
>>Best,
>>
>>Vladimir
>>
>>On Fri, Jun 22, 2012 at 9:30 AM, Elżbieta Olejarczyk
>><[log in to unmask]> wrote:
>>> Dear all,
>>>
>>> Can somebody advise me how resolve this statistical problem?
>>>
>>> I have the following data:
>>>
>>> 20 subjects * 14 experimental conditions * 324 flows (equivalent to one
> set
>>> of 18x18 roi_to-roi) * 5 frequency bands
>>>
>>> I would like to find the answer on the following questions:
>>> 1. which flows are significant for every of the experimental condition;
>>> 2. what is the frequency band for these flows.
>>>
>>> The amount of the data is too big to perform an analysis of variance.
>>>
>>> Could it be possible to treat every set of 18x18 as one map? In such a
> way,
>>> the problem would be reduced to test the following number of maps:
>>>
>>> 20 subjects * 14 conditions * 5 bands
>>>
>>> If this approach is possible, which SPM procedures should be applied and
>>> what format of data should be used?
>>>
>>> Otherwise, what statistical method would be appropriate to resolve this
>>> problem?
>>>
>>>
>>> Best regards,
>>> Elzbieta Olejarczyk
>>
>
|