Re: [Marsbar-users] [SPM] marsbar: percent signal change and contrast value

[Matthew B. <mat...@gm...>]

Hi,

> In one step I extracted the percent signal change of two regressors for the individuals in three groups.
> The second step was the extraction of the contrasts builded through the difference of the two regressors. With this came up some t-stat values.
> Most common in papers is to report the percent signal change for ROIs but the values of the contrast seem to be more convenient for further statistics (especially between groups).
> Now I try to understand the difference between the two kinds of information, because I expected that they are not different. But statistical tests showed significant differences between the difference of the signal change values and the t-stats?
>
> So, could someone be so kind to explain the differences.

Sorry to be slow to reply.   The contrast values and the percent
signal change do differ because the percent signal change values have
been divided by the global mean for the ROI, for the run.  Of course
that can differ across runs and so give different weightings for the
runs.  I guess that's the main cause of what you are seeing.

> And does it make sense to report contrast values or theis t-stats in a paper?

This is something to do with your model of the signal.   Maybe there's
a real change in blood flow R.  You tried to measure R by looking at
the effect on MRI signal, and that caused a measure signal change M,
with a baseline B.   The measured signal change M might well be
proportional to B, so the best way to get something proportional to R
is to divide by B.   I'm happy to be corrected, but my impression is
that most people think M is proportional to B, and therefore I guess
that you'll get closest to something proportional to R - the signal
you want - by dividing by B - and therefore, by using percent signal
change.

Best,

Matthew