Thank you again Karl, but I am still not clear on exactly what gets
plotted.
To get a handle on this, I specified an F-contrast that included only the
constant term. The resulting fitted response looked like the convolved
boxcar, if that was the only other column in the design matrix, or a more
complex waveform if other covariates were included (but not specified in
the contrast). The same result occurred with a t-contrast that included
only the constant term. If I am only testing for the contribution of the
constant term, why isn't the fitted response a flat line? How is the plot
of the model's estimate calculated for a given contrast?
Best regards - Aaron
--
Aaron Field, MD, PhD
Department of Radiology
Wake Forest University School of Medicine
Winston-Salem, NC 27104
Voice: 336-716-9349 Fax: 336-716-2029
Karl Friston wrote:
> Dear Aaron,
>
> > Can someone please tell me, in an fMRI/boxcar analysis with multiple
> > confounds and for a specific contrast, what EXACTLY gets plotted in a
> > fitted/adjusted response (vs. time) plot? I thought this plot showed
> > the estimate that results from the effects specified in the contrast,
> > along with the data adjusted by subtracting the predictions of the
> > confounds. I must have this wrong, because some of the estimated
> > response plots that I have seen cannot possibly result from the
> > specified effects alone.
>
> You are correct that the fitted effects (specified by a contrast) are
> adjusted for the remaining effects in the design matrix. In relation
> to SPM96 the adjustment now considers anything apart from the contrast
> of effects as confounds. By using F contrasts you can now adjust for
> different effects post-estimation. This removal may produce unlikely
> shaped responses. To see the fitted repsonses, without removing
> anything but session effects, select the F contrast for 'all effects of
> interest'.
>
> If this still looks different from a convolved box-car then your
> temporal filtering in probably too severe. The fitted and adjusted
> data are for filtered data. Therefore the fitted responses are
> expressed in terms of filtered regressors. These can have an
> unexpected shape if too many low frequency components have been
> removed.
>
> Although this may be confusing at first we thought it was important to
> show the fitted effects one was actually making an inference about.
> Furthermore the use of this plot facility, in conjunction with
> different contrasts, allows you to dissect different components of the
> response and characterize them graphically.
>
> With best wishes - karl
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|