Hi Jim,
You should either use '-datum float', or multiply your expression
by some cutoff value (or by as mask value, as is done in HowTo #5).
You dataset is probably of type short, and so the sub-bricks are
scaled, where the factor depends on the most extreme value.
Since voxels outside the brain can have huge %change values,
you probably get close to integral resolution of numbers in some
sub-bricks (making it difficult to tell 0 from 0.7, for example).
So using floats gives you the accuracy/numerical resolution, but
the datasets are twice as big.
And using a mask or cutoff value will reduce the tendancy to have
large percent signal changes, so the scale factor will actually do
something useful.
Either way should work out.
- rick