Dear Sahil,

 

I have a very simple enquiry related to your paper "Movement-related effects in fMRI time-series" (http://www.fil.ion.ucl.ac.uk/spm/doc/papers/fMRI_1/fMRI_1.pdf). Do you suggest or is it a good or bad idea to convolve the motion parameters (e.g. head motion of subjects) with HRF also, even if we convolve the data with a Gaussian filter of 8mm FWHM as you mentioned in this paper?

 

Generally, one would not convolve the motion parameters, because their consequences are expressed instantaneously in the data (without delay and dispersion by the HRF).

 

Having said this, if you convolve the data then you have to apply the same operation to the explanatory variables (including the motion parameters). Think of this in terms of a convolution matrix operator W applied to a general linear model

 

Y = X*beta + error  =>

 

W*Y = W*X*beta + W*error

 

The only advantage of applying any W is to make W*error independent and identically distributed over observations – this ensures the ordinary least squares estimates are also the maximum likelihood estimates of beta.  However, SPM already does this for you – so there is no need to apply W.

 

I hope this helps J

 

Karl