Dear Will and Joern
Many thanks for your replies.
It looks as though my interpretation of smoothing betas was in the context
of a complate, multi-session analysis, while Joern appears to be
advocating that this should be done in a partitioned manner at the first-
level(?) I'd just like to clarify the approach for analysing my data (4
sessions, each with 4 ranomized conditions):
- Do four separate, single session analyses wls(ar) on the (spatially
normalized) unsmoothed data at the first-level.
- Review and exclude artefact using the sdscale output from the WLS
toolbox - recompute wls(ar) if data are removed.
- smooth the betas for each of the four separate estimates of each
condition (with four conditions in a single block, this will produce 5
betas, one for each condition plus a block effect, right?).
- take the smoothed betas up to a second-level analysis (two-sample t-
test?)
I assume that to look for contrasts of one condition against the other
across four session, e.g., at the second-level, I would load the betas for
condition 1 and 2 then use a [1 1 1 1 2 2 2 2] vector in SPM to define
them?
Would it also be valid to compute contrasts at the first-level
using "Results", smooth the con image for that contrast for each session,
then take them up to the second-level? If I were to do this in a multi-
subject analysis, would there not be an artificial inflation of df, by
moving 4 measures per subject, as opposed to one, up to the second level?
Apologies for the apparent naivety of these questions!
Regards - MFG
|