Dear Patrick,
I wouldn't expect the residuals mean to be far off zero: which magnitude
are you observing?
The design matrix should contain everything you know about the
experimental design and possible confounds. So you can include
uninteresting conditions - just be careful if they are highly correlated
with effects of interest.
Best regards,
Guillaume.
On 12/10/2019 21:29, Patrick wrote:
> Dear Guillaume,
>
> Thank you for your reply. I took a look at the residuals (and calculated
> their mean over time); several voxels do not have a mean of zero (did
> this by reading the residuals into MATLAB and calculating mean over
> time). Does this indicate a problem?
>
> Also, what about those trivial conditions? Would modeling them lead to
> any problems?
>
>
> Thanks and Regards
> Patrick
> On Wednesday, October 9, 2019, 5:33:30 PM GMT+5:30, Flandin, Guillaume
> <[log in to unmask]> wrote:
>
>
> Dear Patrick,
>
> The residuals should have a mean of zero (over time). To have a look at
> them (assuming you saved them to files using the option in the Model
> Estimation module or used spm_write_residuals), try the following:
>
> * click on ChecReg and select the first residuals image
> * right-click on the image to open the contextual menu and select Browse
> * select all residuals images
> * right-click on the image again and select Browse>Display profile from
> the contextual menu.
>
> Best regards,
> Guillaume.
>
>
> On 02/10/2019 09:10, Patrick wrote:
>> Dear SPM Experts,
>>
>> I am trying to get a better understanding of how to correctly model fMRI
>> data. I have some experimental conditions of interest (say C1, C2, and
>> C3). It is a typical block design experiment where each of the
>> conditions C1, C2, and C3 are preceded by some instructions (lasting 2
>> TRs). I am currently modeling the instructions as a separate condition.
>> My question is: is this a good idea? Or is modeling trivial incidents in
>> the experiment not going to add anything meaningful to the model but
>> will only constrain the model? I use the word trivial to indicate any
>> such condition like instructions which are never really going to be of
>> interest and last for a very small duration.
>>
>> Second part of the question is that when I take a look at the residuals
>> after model fitting, I see that most of the voxels have negative values.
>> Does this indicate a flaw in the model...perhaps I am overfitting?
>>
>>
>> Thank you in advance for any insight into this
>>
>>
>> Best Regards
>> PB
>
>
> --
> Guillaume Flandin, PhD
> Wellcome Centre for Human Neuroimaging
> UCL Queen Square Institute of Neurology
> London WC1N 3BG
>
--
Guillaume Flandin, PhD
Wellcome Centre for Human Neuroimaging
UCL Queen Square Institute of Neurology
London WC1N 3BG
|