Dear Liam,
> I am currently running higher level (between groups unpaired t-tests)
> analyses and I have a query regarding threshold values.
>
> FSL has the Z value threshold set to 2.3 by default. My understand
> is that
> 2.3 is the equivalent to 0.01 one-tailed and 1.96 is the equivalent
> to 0.05
> two-tailed. Is this correct?
As Steve indicated these are thresholds used to form clusters. A
cluster is a contiguous set of connected voxels. A given cluster will
have a "size", i.e. the number of connected voxels. The inference is
based on this size, where the logic is that a large cluster is
unlikely to have occurred by chance (i.e. if the null-hypothesis is
true).
The "cluster forming threshold" is arbitrary, and you can use pretty
much any value greater than ~1.6 (which I think is the rule of thumb
lower limit where some of the assumptions start breaking down). Just
remember that the smaller the initial cluster forming threshold is
set, the larger the cluster must be to reach significance. This means
that the clusters you find with small cluster forming thresholds tend
to be large, and hence difficult to interpret since they may extend
across several functionally distinct areas.
A larger cluster forming threshold will often reduce your sensitivity,
but may yield smaller clusters that are easier to interpret.
Good Luck Jesper
|