Hi,
On 28 Jan 2010, at 09:10, Silvia Juanes wrote:
> Hi,
>
> We have some questions as regards the randomise statistical
> procedure and we will be very grateful if someone could shed some
> light on these:
>
> 1. Is the 'results_tstat1.nii.gz' image composed by statistics that
> follow a t-student distribution or, on the other hand, correspond to
> an empirical distribution created on the basis of the null
> hypothesis by using the permutation process, being the contrasts
> tested from the accumulated probability?
It's the raw "tstats" output by the GLM at the heart of randomise and
so generally will follow a t distribution - although part of the point
of the permutation based conversion of these values into p-values is
that randomise will get that conversion correct regardless of the
actual null distribution of this t test statistic.
> 2. These t values correspond to p-values of the
> results_vox_tstat1.nii.gz image? and how can we know the
> corresponding t-value of a cluster corrected p-value (of the image
> results_maxc_tstat1.nii.gz)?
I think you mean "_vox_p_tstat" (if not you should upgrade to the
latest version of FSL) and yes these are the uncorrected p-values
corresponding to the t values discussed above. There is no direct
mapping between voxel t values and cluster p-values: the logic goes:
form raw t, threshold to form clusters, decide on some cluster test
statistic (like size), and feed that into the permutation-based null
distribution generation so that size can be converted into cluster p-
values.
> 3. As regards the degrees of freedom of these analyses. Do they
> really involve some degrees of freedom? how can they be calculated?
> does it has a coincidence with 'classical' statistical rules?
These are not explicitly needed - that's dealt with implicitly via the
permutation testing.
> 4. Finally, in the case we want to compute a r statistic. Would it
> be correct to transform the t value of that same coordinate , or on
> the other hand should be compute it from the corresponding p-
> value(specially if the answer to the first question is that the
> outputs are not t-students)?
If you want a voxelwise r then yes compute it from the t is probably
the easiest thing.
Cheers.
>
>
> Thanks a lot in advance,
>
> Silvia Juanes
---------------------------------------------------------------------------
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
---------------------------------------------------------------------------
|