Eleonora
I am assuming your A, B and rest (R) conditions are blocked and modelled with
a simple boxcar.
Firstly, because you have specified A, B and R as separate conditions in your
Fixed Effects models, you will not be able to estimate parameters for each
condition uniquely - the covariate for the R condition, for example, is simply
a linear combination of those for the A and B conditions.
You should perform another analysis where the model contains only conditions
A and B (and in which a [1 0] implicitly compares condition A against rest).
You
can then take the resulting con*.imgs through to a second-level analysis.
To compare conditions A and B against rest in a Random Effects analysis across
subjects,
you would simply take the 6 con*.img images for [1 0] or [0 1] contrasts in the
Fixed Effects analyses, and enter them into a one-sample t-test. To compare
conditions A against B directly,
you could either enter the 2x6 relevant con*.img images into a paired
two-sample t-test,
or equivalently, construct [1 -1] contrasts in the Fixed Effects level, and
take these 6
con*.img images into a one-sample t-test.
Note that even if you could estimate A, B, and R conditions uniquely, you
should avoid
taking more than two conditions (per factor) through to a second-level analysis
(such
as the PET analysis you suggest), because SPM makes no correction for
nonsphericity
of data at the second-level.
Note also that a similar procedure would apply to comparing sessions within a
condition,
or if you want to collapse across sessions of each condition, the relevant
Fixed Effects
contrasts would be [1 0 1 0] or [0 1 0 1] for A vs R and B vs R, for example.
Note finally that you should not expect high t-values in your Random Effects
analysis
with only 6 subjects!
Hope this helps,
Rik
> I have performed a fixed effect model for each subject so I have the
> Con_*.img files corresponding to the effects of interest to introduce in a
> second level statistic but, first of all, I don't know which is the best
> model to analyze
> 6 subjects
> 2 session/subj
> 2 condition + baseline
> 1scan/condition
> Can I also use PET model or not (for example considering the problem like:
> 2 groups, 6subj/group, 2cond+Rest, 1scan/cond)?
> After the theoretical problem, there is the practical one:
> I tried to specify some statistical random effects model only to understand
> how to work but I have always the same problem: the 'beta' is not uniquely
> specified (I don't know if I have to consider Rest state or not and, if
> yes, how).
> In the same way I tried using adjMean output files instead of Con_*.img
> files for each subject but the problem is still there.
>
> Then...there is someone who can help me and explain, step by step, in a
> simple and practical way, what I should do?
>
> Thanks a lot for spending this time.
>
> Eleonora Fornari
--
---------------------------8-{)}-------------------------
DR R HENSON EMAIL [log in to unmask]
Wellcome Department of
Cognitive Neurology TEL (work1) +44 171 833 7483
12 Queen Square TEL (work2) +44 171 833 7472
London, WC1N 3BG FAX +44 171 813 1420
URL: http://www.psychol.ucl.ac.uk/rik.henson/index.html
---------------------------------------------------------
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|