Hi,
On 4 Jul 2007, at 18:19, Mara Cercignani wrote:
> HI Steve,
> Thanks, but let me get this straigh:
>
> 1) Is the orthgonalisation really necessary?
In "normal" GLM, no it isn't, but in such scenarios in permutation
testing, yes it is. This is because you can't leave the regressors in
as confound EVs in the main matrix, because then the permutation
doesn't make sense - you need to pre-regress the EVs out - in which
case you then need the orthogonalisation.
In the new version it will be cleverer - for each contrast an
effective regressor will be formed, along with an orthogonal confound
matrix, but this will all be done automatically for you.
> 2) Do all the variables need to be orthogonal (eg is testing
> regressor 1,
> and using regressors 2 and 3 as confounding factors, do regessors 2
> and 3
> need to be orthogonal as well?), or just 1 and 2 & 1 and 3?
The confounds don't need to be orthogonal to each other, but if you
are regressing them out of regressor of interest one at a time they
DO need to be, otherwise in general after the second
orthogonalisation the orthogonalised regressor is no longer still
orthogonal to the first confound.
Cheers.
>
>
>> Hi Mara,
>>
>> Hi - you can, but you need to be very careful about the details. To
>> be as accurate as possible, you would need to do a separate analysis
>> for each of these regressors, orthogonalising the regressor of
>> interest wrt the "confounds" (ie the others in the model) and putting
>> them into the -x matrix. E.g. orth R1 wrt R2 and the interaction,
>> then put R1 and R2 into the -x matrix.
>>
>> Or - if you wait a few weeks for the new release, the new version of
>> randomise will do all this for you!
>>
>> Hope this helps? Cheers, 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
------------------------------------------------------------------------
---
|