Hello Sung,
Oops... forgot to add that to get proper % change you also need to
divide the value I mentioned in previous e-mail by the baseline
signal (approximated well in BOLD fMRI by the overall mean,
estimated by the "session effect" beta value) and multiply by 100.
Let me know if any of this is unclear.
Eric
Quoting Sung Lai Yuen <[log in to unmask]>:
> Dear Matthew,
>
> Thanks for your answer! What I want to follow-up is....if I'm
> using a simple
> HRF model, is it justified for me to report the contrast
> estimates *as* the
> estimated % signal change of contrast A-B in a paper? Are there
> any
> references for me to study this issue in more detail?
>
>
> Thanks again,
> Kenneth Yuen
>
> On 2/21/07, Matthew Brett <[log in to unmask]> wrote:
> >
> > Hi,
> >
> > > Sorry if this is more a MarsBar-related question: I read from
> the
> > MarsBar
> > > discussion list that if I got a contrast A-B (1 -1) and wanna
> get the
> > > respective %sc for the contrast, I have to find the %sc for
> condition A
> > and
> > > %sc for condition B individually and subtract them manually.
> What I
> > would
> > > like to know is...how is the differences in %sc between
> conditon A and B
> > > related to the contrast estimates of the contrast A-B?
> >
> > If you have just an HRF model of your events, A-B will be very
> similar
> > in proportion for the contrast and the event % signal change.
> As the
> > model becomes more flexible, the events can change shape, and
> to the
> > extent they do so, the contrast of - for example - simple HRF
> > regressors will become more and more different from the %
> signal
> > change measure...
> >
> > Best,
> >
> > Matthew
> >
>
|