Dear Aaron
> I know that autoregressive-moving-average (ARMA) modeling has been used
> in SPM as a preprocessing step, to adjust fMR data for motion effects.
> Does anyone incorporate an AR model into the general linear model
> during the regression stage, i.e., enter lagged versions of the fMR
> time series into the design matrix? This seems like an obvious thing to
> do, not so much as a motion correction but as a better model for the
> data in general. I haven't seen any discussion of this, and I'm a bit
> curious...
Not explicitly (although there is an AR(1) model for the seriel
correlations in the residuals). Given that any AR process has an
equivalent moving average (i.e. convolution) formulation the
convolution of the underlying stimulus functions by basis functions can
be considered from the point of view of ARMA (here the innovations are
the stimulus functions). Ed Bullmore and Mick Brammer use an AR
pre-whitening strategy to remove serial correlations prior to model
fitting but this is not an approach adopted in SPM.
With best wishes - Karl
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|