Hi Amit,
If you are not interested directly in asking about the baseline, I would
suggest you to use implicit baseline. If you have one only task in each
session with a block design (two tasks, one per session), you can look at
the attached image to have an idea. As far as I'm concerned, exlicitly
modeling the baseline will decrease the statistical power because you are
adding parameters to the model (empirically this means there is less
activation in areas where I expect activation, in my functional studies).
Hope this helps,
Juan J. Lull
______________________________________________
Juan J. Lull - jualulno_at_upvnet.upv.es
[MI - Medical Imaging Area]
BET - Bioengineering, Electronics and Telemedicine Group
UPV - Politechnical University of Valencia - Spain
http://ttt.upv.es/jualulno
______________________________________________
----- Original Message -----
From: "Amit Etkin" <[log in to unmask]>
To: <[log in to unmask]>
Sent: Tuesday, October 12, 2004 6:02 PM
Subject: Re: [SPM] 1 1st level model w/ 2 sessions or 2 1st level models
with 1 session
> If you include 2 sessions in a single model, how would you deal with the
baseline? Would you
> implicitly model it (ie only model the activity periods), or would you
explicitly model a
> separate baseline regressor for each session? For the comparisons I'm
interested in I contrast
> an activity of interest condition to a control activity condition, but
would like to get out main
> effect contrast estimates.
>
> thanks,
> Amit
>
>
> >> I am running an experiment with two related tasks in a group of
normals. My question is
> this:
> >> is there any reason why I should make two separate 1st level models for
each subject (one
> per
> >> subject), rather than one model with two sessions (one for each task)?
Will running one
> bigger
> >> model change the way effects and variances are estimated? or influence
my ability to
> compare
> >> the magnitude of activation in a given brain region between the two
taks.
> >>
> >
> >
> >Re. the effect estimation and variance estimation issues - both will be
estimated differently
> >in the two model versus one model approach. Which approach is better
depends on how
> closely
> >matched the two sessions are. The one model approach has more data to
explain but has
> more
> >parameters to play with. Whether or not the error variance (assumed
constant over both
> sessions) is
> >higher or lower is an empirical issue.
> >
> >However, from a flexibility perspective, I think running one big model
(ie. modelling both
> sessions
> >in the one model) is the better option here.
> >
> >This is because it allows you to assess differences between sessions.
You'd certainly
> >want to do this if an experimental effect of interest was expressed over
sessions eg.
> >session 1: before drug, session 2: after drug.
> >
> >Then, you could use the contrast manager to estimate the drug effect
> >in each subject.
> >
> >Without doing this there is no easy way in SPM to assess between session
(within-subject)
> >effects.
> >
> >Best,
> >
> >Will.
> \
>
|