Dear Silvia,
> Let's say I have an event-related design in which I am only interested in a
> portion of each trial (e.g. scans 5 through 10 of a 20-scan trial). If I
> select the event-related model, I can specify only the starting scan number
> for each trial (which in this case would presumably be the 5th scan of each
> trial). On the other hand, if I select 'epoch', I get to specify not only
> the onset but also the duration of each event of interest (in this case, 5
> scans). Which way should I do it?
The answer depends on what your expectations are about the form of the
neural activity during scans 5-10. Using 'epoch' you are assuming that the
mean activity during this period is well fit by a box car (i.e. roughly
constant level of activity throughout the period). Using 'event' the
expected form of the activity will depend on the basis functions you select.
Selecting hrf will assume that the activity between scans 5-10 is well fit
by a delta function at scan 5 convolved by the hrf; whereas selecting (e.g.)
fourier basis will give a (much) less constrained fit.
So alas there is no generic recommendation; it depends on your experiment.
In general the basis you select should reflect the expected neural activity
during the period modelled (so it's a good idea to have a reasonable
expectation of what the form of that neural activity will be!). A well
fitting model (no systematic structure for the residuals) is a prerequisite
for reliable statistical estimation.
best wishes,
Geraint
--
Dr. Geraint Rees
Wellcome Advanced Fellow, Lecturer,
Division of Biology 139-74, Institute of Neurology,
California Institute of Technology, University College London,
Pasadena CA 91125 London WC1N 3BG
voice 626-395-2880 020-7833-7472
fax 626-796-8876 020-7813-1420
web http://www.klab.caltech.edu/~geraint
--
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|