Ian -
Just to follow up one point in Richard's excellent response.
> >one alternative that we have tried is to start with the fixed effects
> >analysis, and then followup by asking if the regions identified in the fixed
> >effects are reliable across the subjects in the sample using roi extraction
> >and random effects anova on the extracted timecourse data for each subject.
>
> Sorry, I didn't quite follow this. Perhaps someone else will comment. But...
>
> >if the condition effects and condition by time interactions are significant,
> >one can assume the activation is reliable in the sample.
Putting binned data, averaged across subjects, into a repeated measures
ANOVA is a satisfactory random effects analysis, but you must correct
for sphericity violations (the data from each time point are unlikely to be
i.i.d for example). This is why you are advised to stick to 1 or 2 sample
t-tests in SPM99, for which there is no sphericity issue. Karl and others
are currently working on sphericity corrections for the next release of SPM,
which will allow more complex factorial second-level designs.
We have already implemented a Finite Impulse Response (FIR) basis set in
the SPM version under development (think of as a set of mini boxcars at
each peristimulus time), for which the parameter estimates will correspond
exactly to the binned data you are talking about (so you can do the procedure
you suggest in the usual manner of taking through con*imgs, without needing
to extract any data by hand).
More importantly however, though a condition x time interaction in your
ANOVA might indicate a significant event-related response (without making
any assumptions about the shape of the response), you will find it MUCH
less powerful than a t-test on the parameter estimate for the canonical HRF
(if the response does resemble a "normal" HRF). The question then becomes
how often are HRFs non-canonical, or more precisely, what proportion of
variability is missed by the canonical HRF. In simple event-related designs
where you believe the neural activity is short-lived and immediately follows
the stimulus (ie adequately modelled by a delta function), I would suggest
not much variability is missed (data to be presented at HBM 2001). Of
course, if you expect more complex patterns of neural activity (eg delayed
by >1s, or prolonged during some delay period), then convolving delta
functions by a canonical HRF alone will not be sufficient. In this case however,
I would suggest that you revise your underlying neural model (ie delta functions
are not sufficient, and you should model delays for example by mini box-cars).
(Alternatively, if you have less idea about the underlying neural model, you might
use Eric's recent idea of assuming a fixed form for the HRF, but a set of basis
functions to capture a range of different underlying neural models.)
Rik
--
---------------------------8-{)}-------------------------
DR R HENSON
Institute of Cognitive Neuroscience &
Wellcome Department of Cognitive Neurology
17 Queen Square
London, WC1N 3AR
England
EMAIL: [log in to unmask]
URL: http://www.fil.ion.ucl.ac.uk/~rhenson
TEL1 +44 (0)20 7679 1131
TEL2 +44 (0)20 7833 7472
FAX +44 (0)20 7813 1420
---------------------------------------------------------
--
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|