Hi - voxelwise
Cheers
--------------------
Stephen M. Smith, Professor of Biomedical Engineering
Associate Director, Oxford University FMRIB Centre
FMRIB, JR Hospital, Headington,
Oxford. OX3 9 DU, UK
+44 (0) 1865 222726 (fax 222717)
[log in to unmask]
http://www.fmrib.ox.ac.uk/~steve
----------------------
On 13 Mar 2010, at 20:14, Michael Harms <[log in to unmask]> wrote:
> Ok. For some reason I thought that the "vox_p" values were based on
> the
> parametric t->p transformation. So, for the "vox_p" values, is the
> empirical distribution calculated separately at each voxel, or is
> there
> just a single empirical distribution formed by aggregating values from
> all voxels within the mask that is then used to compute the
> empirical p-
> value at each individual voxel?
>
> thanks,
> -MH
>
> On Sat, 2010-03-13 at 09:45 +0000, Stephen Smith wrote:
>> You are comparing the parametric t->p transformation (using the
>> theoretical t distribution) with the nonparametric transformation
>> (derived by permutation testing that estimates the _actual_ null
>> distribution of the test statistic - here t). The whole point of
>> randomise is that it estimates this transformation empirically (from
>> the data) - so it won't be the same as the theoretical transformation
>> unless your data+model _exactly_ follow the parametric distributional
>> assumptions. For a large enough number of permutations the empirical
>> approach is guaranteed to be more accurate.
>> Cheers, Steve.
>>
>>
>>
>> On 12 Mar 2010, at 21:53, Michael Harms wrote:
>>
>>> Hello,
>>> My understanding is that the "_vox_p_tstat1" output of randomise
>>> (from
>>> the -x flag) reflects the p-value calculated according to a t-
>>> distribution, where the t-value itself is contained in the
>>> "_tstat1.nii.gz" file.
>>>
>>> However, the two values do not always seem to match as I would
>>> expect,
>>> and to an extent that appears to be more than just "rounding error"
>>> related to whatever library is used to compute statistical
>>> distributions.
>>>
>>> For example, I'm looking at a voxel (using FSLView) for which the t-
>>> value is 1.51444, and the p-value, according to "_vox_p" image is
>>> 0.8818. This is from an analysis having 150 volumes, and 5 columns
>>> in
>>> the design matrix -- i.e., 145 dof.
>>>
>>> According to Matlab the p-value for that distribution is:
>>>>> tcdf(1.51444,145)
>>> ans =
>>> 0.933954466613827
>>>
>>> which I consider a rather "large" discrepancy from the "_vox_p"
>>> value of
>>> 0.8818. Any idea as to the source of these difference?
>>>
>>> thanks,
>>> -Mike H.
>>>
>>> --
>>> Michael Harms, Ph.D.
>>> --------------------------------------------------------------------
>>> Conte Center for the Neuroscience of Mental Disorders
>>> Washington University School of Medicine
>>> Department of Psychiatry, Box 8134
>>> Renard Hospital, Room 6604 Tel: 314-747-6173
>>> 660 South Euclid Ave. Fax: 314-747-2182
>>> St. Louis, MO 63110 Email: [log in to unmask]
>>> --------------------------------------------------------------------
>>>
>>>
>>
>>
>> ---
>> ---
>> ---------------------------------------------------------------------
>> 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
>> ---
>> ---
>> ---------------------------------------------------------------------
>>
>>
>>
>>
>>
>>
>
|