Daniel -
It's a tricky question...
I would suggest option B. If you only had 2 event-types with cues
only, the results from option A and option B should be identical -
one model would simply be a reparametrisation of the other.
Given that you have targets as well for event-type I, which occur
slightly later than the cue, option B may be a more accurate model
than option A with only one (cue) onset - though I would expect the
differences to be small, given that two events 1.5s apart, when
convolved with an HRF, will produce a joint HRF that is probably
very similar to a scaled version of the HRF (assuming linearity).
In this case, a larger parameter estimate for event-type I than II
in option A would reflect the influence of the target. Moreover,
(temporal and) dispersion derivatives should indeed soak up at
least some of the residual shape differences.
I'm not sure, but I think the inclusion of two onsets in option A, one
for cue and one for target, within the same event-type, is not suitable,
because while it might capture the shape of the joint response slightly
better, it assumes that the response to cue and target are of equal
magnitude, which may not be the case. (What is less advisable
would be to model the cue and target of event-type I as separate
event-types - though a more general model, the cue and target
regressors will be highly colinear, meaning that separate tests on
each one alone will have little power, though F-tests on both
would be okay).
Rik
> Dear SPMers,
>
> I'm trying to decide on the best way to model an event-related response
> and thought I'd seek your input. In my experiment, there are 2 basic types
> of randomized events. In Event Type 1, a cue stimulus is presented for 200
> ms and then followed by a target stimulus 1500 ms later. In Event Type 2,
> the same cue stimulus is presented for 200 ms, but no target is ever
> presented. Each type of event lasts 3 seconds and my TR is 1.5.
>
> I'd be interested to know what people think about the two options below
> for modelling these events.
>
> (A) Model event types I and II with different regressors
>
> (B) Model all cue stimuli with one regressor and target stimuli with a
> second regressor.
>
> Option A seems very straightforward to me. One issue here is whether to
> address the fact that event type I has 2 stimuli by (1) including a
> dispersion derivative or (2) specifying two onsets for event type 1 - one
> for the cue, one for the target. In practice, specifying two onsets seems
> to better capture target-related activity than does using a dispersion
> derivative. But, is specifying two onsets the best approach? It seems to
> assume that cue and target stimuli will produce identical responses.
>
> Option B seems tenable as well since the same cue stimulus is presented
> in Event types I and II. I worry, though, that there might be a
> multicollinearity problem because all target stimuli are always preceded by
> the same cue stimulus. Still, not all cue stimuli are followed by a target
> (66% of cues are followed by a target, 33% are not). In this situation, can
> SPM come up with independent estimates of the responses to cue and target
> stimuli? In practice, this approach seems to do a better job of capturing
> cue-related activity than Option A, although target-related activity is
> slightly weaker, which made me think of the multicollinearity issue.
>
> Any advice would be much appreciated!
>
> Thanks,
>
> :> Daniel
>
> Daniel Weissman, PhD
> Center for Cognitive Neuroscience
> Duke University
> Durham, NC 27705
> phone: (919)-681-1029
> fax: (919)-681-0815
> e-mail: [log in to unmask]
--
---------------------------8-{)}-------------------------
DR R HENSON
Wellcome Department of Cognitive Neurology
& Institute of Cognitive Neuroscience
University College London
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
MOB +44 (0)794 1377 345
FAX +44 (0)20 7813 1420
---------------------------------------------------------
--
|
