One thing which may be relevant is I believe the SPM99 SPM regressors were
zero-mean, whereas from SPM2 onward they are not zero-mean
(http://www.jiscmail.ac.uk/cgi-bin/wa.exe?A2=ind04&L=SPM&P=R65537&I=-3).
There are good reasons for this, but it could be why you are seeing what you
see.
VDC
> -----Original Message-----
> From: SPM (Statistical Parametric Mapping)
> [mailto:[log in to unmask]] On Behalf Of Scott Slotnick
> Sent: Thursday, September 21, 2006 1:27 PM
> To: [log in to unmask]
> Subject: [SPM] Negative SPM5 beta-weights
>
> Hello,
>
> I recently conducted an analysis of a dataset in SPM5 that I
> had previously analyzed in SPM99 (with standard random effect
> GLMs). On the plus side, the pattern of SPM activity was
> identical, the only difference being slightly more robust
> responses with SPM5. On the minus side (figuratively and
> literally), the SPM5 beta-weight values extracted from visual
> activity evoked by left and right hemifield stimuli were
> almost all negative in amplitude (the same beta-weight values
> were almost all positive in SPM99, as would be expected from
> retinotopic increases in neural activity). To follow-up on
> this, I plotted a histogram of each beta-weight set
> (beta_0001.img, beta_0002.img, etc.) resulting from the
> analyses. Values of most SPM5 beta-weight sets had a
> relatively large baseline shift (some positive, some
> negative, often skewed in distribution), while values of all
> SPM99 beta-weight sets were centered at 0 with far less
> variance and no skew (as would be expected, given that only a
> small fraction of voxels should be associated with a given
> event type). What is interesting is that the relative
> difference in magnitude between beta-weights seems to be
> intact, explaining why the overall activation pattern for
> both analyses is the same.
> For instance, in left extrastriate cortex the beta-weight
> value corresponding to right visual field stimulation is
> greater than the value associated with left visual field
> stimulation for both analyses, but the mean beta-weight value
> is significantly positive for the SPM99 analysis and
> significantly negative for the SPM5 analysis. Finally, I
> should also mention that I'm interested in this because I
> have written a function that extracts beta-weights from ROIs
> to reveal the sign of event-related activity, and while it
> had been useful in SPM 99, it is currently useless in SPM5
> for the above reasons. Any ideas would be greatly appreciated.
>
> Thanks,
> Scott
>
|