Dear Darren,
> I'm running a set of analyses in which I'm removing certain covariates
> of no interest from a dataset- specifically movement parameters and a
> set of sign and cosign function which effect a low-pass filter.
>
> When I run the analysis in spm97 I get about a 100 dB signal reduction
> in the high frequencies above the filter cut-off point (assessed by
> looking at the power spectrum of the adjusted data); however in spm99
> the same covariates produce only about a 20 dB reduction.
>
> Why? and how can I get better results out of spm99?
In principle the results should be the same. The only thing I can
think of is that data in Y.mad have less prescision than XA.mat (due to a
more efficient compression). It may be that rounding errors contribute
to your high frequencies. You could try applying the residual forming matrix
to these data directly by replacing line 170 with
% residuals
%---------------------------------------------------------------
R = spm_sp('r',xX.xKXs,spm_filter('apply',xX.K,y));
> I've looked at the covariates in the G matrix for SPM96/7 and xX.X for
> spm99 and they look exactly the same, so I'm guessing this has
> something to do with how the convolution matrix is implemented
> differently. But that's just a guess.
The low frequency regressors are now in:
xX.K{s}.KH
for session s. and the Low-pass convolution kernel is in
xX.K{s}.KL
but I do not see how this explains the discrepancy. Let us know if you
sort it out.
With best wishes - Karl
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|