Hi Ahmed,
> I am comparing 2 conditions within one subject with multiple
> replications (actually I am trying to statistically compare the raw
> connectivity maps derived from a diffusion tractography algorithm).
>
> There are 30 replications.
>
> It prompts me for size of exchangeability block. What does this mean
> and what could be the optimum choice for me?
The exchangeability blocks (EB) are a way to account for temporal
confounds in PET or other slow-measurement methods (their use with
strongly autocorrelated data (fMRI) is still under study).
The fundamental assumption for SnPM is exchangeability under the null
hypothesis. Many SnPM plugins simply assume that all observations
are exchangeable, and don't ask about exchangeability blocks (EBs).
For example, with a one-sample t test (single group, one scan per subject)
EB's are never mentioned; this is because independence between individuals
justifies exchangeability (the EB size the whole dataset).
I don't know about the drift concerns in diffusion tractography.
If there are none you can specify a EB size of the whole dataset (30,
or is it 60?). This will give you 60-choose-30 or 1.2 x 10^17 possible
permutations, an ample number to choose from.
If you have concerns about temporal drift over your replications,
you can specify a smaller size. The number of permutations is
(EBs-choose-EBs/2)^nEB, where EBs is EB size and nEB is the number
of EB's. For example, if you collected the data ABABABAB..., you could
have 30 EB's of size 2 (EBs=2, nEB= 30) and (2-choose-1)^30 or just over
a billion permutations.
In summary: If you have lots of replications, you can choose a small EB
size to be safe. If you don't have much data, though, small EB's
will lead to too few permutations (1000-10000 are desireable).
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
|