Dear Boris,
>we did a study with the followng conditions
>
>movement right hand without medication
>movement left hand without medication
>movement right hand with medication
>movement left hand with medication
>
>in sum 240 scans, d.h.:
>
>10 rest 10 activ 10 rest 10 activ 10 rest 10 activ per condition:
>4 x 60!
>
>Also scans per session = 240
Mmm. It sounds as though you have gone through this procedure for
right hand without medication (60 scans) then left hand without
medication (60 scans), then right hand with medication (60 scans),
then left hand with medication (60 scans). You should probably
therefore consider each 60 scan block to be a 'session'. In effect
this means that SPM will add a 'session' effect, a column of ones
which models any difference in the average signal from one session to
another.
So, when you are prompted for the number of sessions you would now
say '4'. Or scans per session would be '60 60 60 60'. 'Are
conditions replicated?' would need the answer 'yes', and 'Are
timing/parameters the same?' also needs 'Yes'.
>So how do I have to specify the modell
>
>number of conditions = 4 ???????? (Whats with the rest condition???)
It is reasonable to model your rest condition 'implicitly', (i.e. as
a linear combination of other regressors rather than with its own
regressor). If you adopt the approach of considering each experiment
as four sessions, then you only need to have one 'condition' (which
you are now comparing between sessions).
From what comes next I presume that you have chosen 'fixed' as the
SOA option...
>SOA Scans for active 20 ???????? (there a in sum 60 scans...?????)
>
>First trial scan = 10 ????
>epoch length for active scan = 10 ????
These should now all work as you intend (and you'll only have to
enter them once, for the one trial type).
(Incidentally the reason why they didn't before is that SPM assumed
that after the first trial scan (at scan 10), there were trial scans
at 20 scan intervals throughout the rest of the experiment (240
scans), which is not what you intended.)
Incidentally, if you really don't want SPM to add a separate session
effect for each 60-scan session, then you'll have to answer
'Variable' to the SOA and then put in a vector of onsets. I think
that it's better to leave the session effects in the model, though.
>I tried a lot, but it doesn' t work in a correct way, so many thanks for
>the help!!!
No problem. I can't help commenting that there may be a better
design for this experiment. I can understand that you might feel
that you have to do 'no-drug' before 'drug' every time, but it does
mean that the effect of the medication is confounded by time (or
repetition, or session, or whatever). I assume that it is not
feasible to test half of the subjects with 'drug' followed by
'no-drug', in which case the reviewers will accept this.
However, why not interleave left hand and right hand responses? This
way at least the effect of hand is not confounded by time. So if you
find that an area appears to show a lateralized response (e.g. bigger
response to right hand than to left), you'll know that this isn't
really just an effect of time (e.g. bigger response earlier in the
experiment than later in the experiment). If this isn't feasible for
some reason, one could at least have half subjects with the left hand
tested first and half with the right hand tested first.
Best of luck,
Richard.
--
from: Dr Richard Perry,
Clinical Lecturer, Wellcome Department of Cognitive Neurology,
Institute of Neurology, Darwin Building, University College London,
Gower Street, London WC1E 6BT.
Tel: 0207 679 2187; e mail: [log in to unmask]
|