Esa,
> We have a PET study with 8 subjects and two scans per subject, with 7
> degrees of freedom in the analysis. How do the low degrees of freedom
> affect the statistics of this design using SPM?
The random field theory behind corrected p-values for t images is
conservative for low degrees of freedom [1]. At very low degrees of
freedom, smoothness is too low, which will also cause problems
(conservative p-values).
And, of course, as with any statistical method, with less data you have
less power to detect a signal.
> Is SnPM particularly suitable analysis framework in this case, or do
> the low degrees of freedom hamper it as well?
Yes, SnPM is particularly useful in the low df case. First, since
it doesn't use random field theory, it doesn't exhibit the
conservativeness mentioned above. Further, if you use variance
smoothing to create "pseudo t" images, you effectively increase
your degrees of freedom and increase your power.
Andrew Holmes and I have written a primer on SnPM that is to appear in
the January issue of HBM; it's on line now:
http://www3.interscience.wiley.com/cgi-bin/issuetoc?ID=86010643
In particular, see the third worked example for a comparison of
SPM and SnPM under low df.
> Is SnPM equally appropriate for a split-plot design (2 groups with 8
> subjects each; two scans per subject)?
Yes; assuming that each of the two scans per subject are different
conditions, use the following design:
'MultiGroup: 2 groups, 2 conditions, 1 scan per condition',...
If the two scans are actually the same condition, I would collapse them
with AdjMean then use the 'MultiGroup: 2 groups, 1 scan per subject'
design.
Hope this helps.
-Tom
-- Thomas Nichols -------------------- Department of Biostatistics
http://www.sph.umich.edu/~nichols University of Michigan
[log in to unmask] 1420 Washington Heights
-------------------------------------- Ann Arbor, MI 48109-2029
[1] Stoeckl J, Poline J-B, Malandain G, Ayache N, Darcourt J (2001):
Smoothness and degrees of freedom restrictions when using SPM99.
NeuroImage 13: S259.
|