As far as I recall, SPM.xX.W contains the estimate of the
autoregressive model, so it is not directly manipulable. I'd rather
create my own weighting matrix and multiply it by SPM.xX.W, but I'd
worry about whether this really fits in with how SPM handles the code.
The method of adding an extra regressor results in excluding the
volume from the estimates, so it's a clean alternative. It is called
Bartlett's ANCOVA method and you may find more technical details on it
in Little & Rubin, Statistical Analysis with Missing Data, Wiley, §2.5
of the second edition.
Best wishes,
Roberto Viviani
University of Ulm, Germany
> I'd like an answer to exactly the same question in a different context.
> I've got NIRS data from babies with some artefacts and I need to model it
> in SPM but exclude the artefacts. I'm pretty sure it is possible to do
> this using the SPM.xX.W matrix but I cannot work out how from the code.
> And I don't want to just change things haphazardly.
>
> Can any of the SPM gurus give us some guidance on this weighting matrix?
>
> I know I could use the approach Donald McLaren suggested of adding an extra
> regressor in my design matrix to model out the artefact, but I'm not so
> confident in that. Can that really soak up all my horrible spiky
> artefacts? Or will it just reduce their contribution?
>
> best
>
> Antonia
>
>
>
>
> On 27 September 2011 09:29, Eelco van Dongen
> <[log in to unmask]>wrote:
>
>> Dear SPM list,
>>
>> I am currently analyzing long scanning sessions obtained during sleep (2-3
>> hours in length) of which only a part (+- 30min) is of interest. Because of
>> the special nature of the recordings, noise and artifacts due to motion
>> have a big influence on the data quality.
>>
>> As far as I understand, it should be possible to instruct SPM to weight
>> the relevance of individual volumes for estimation with the SPM.xX.W
>> matrix. It would be great if I could use this approach to discount
>> artifact-rich and irrelevant scanning periods to minimize the impact of
>> motion and other noise in my analysis without disturbing the temporal
>> sequence of the scans.
>>
>> Looking at the contents of this variable in one of my normal 1st level
>> design matrices, I am confused how to change it to suit my purposes. From
>> my relative ignorant perspective, it seems it is structured as a Nvolumes x
>> Nvolumes matrix containing values from -0.16 to 1.08.
>>
>> I would be very grateful if somebody could offer me some input on:
>>
>> A: How the weighting matrix is structured
>> B: To what value I should change the weighting for those volumes I want to
>> disregard in the model estimation
>> C: Whether this is a valid approach to take, considering the circumstances
>>
>> Many thanks,
>>
>> Eelco
>>
>
|