Hi Gonzalo,
With a cluster statistic, the choice of height (Z) threshold is
arbitrary. Lower Z values mean that significant extents will be
larger. So for a Z threshold of 2.3, you tend to get fairly large
extents that are good for identifying large systems like cortical
motor areas, or visual cortex, or similar. Higher Z thresholds
(Z=3.1, 3.5, or even 4.0) are better for identifying more focal
activations.
There is a theoretical range for height values. Too high and they
become exactly the same as a voxel statistic -- that is, when the
chance of a single voxel extent is 0.05 then there is no point going
any higher. I seem to recall there is a lower limit as well, beyond
which the GRF assumptions start to break down. I can't imagine ever
really needing to go much lower than 2.3, though, but perhaps someone
else can chime in to explain why even lower thresholds might be useful.
Hope this helps.
Joe
Joseph T. Devlin, Ph.D.
Cognitive, Perceptual & Brain Sciences
University College London
Gower Street
London, WC1E 6BT
email: [log in to unmask]
|