If you want to make sure you have floating point data values
with 3dcalc, you need to use the "-datum float" option. I'm
guessing that you used "-fscale", which does not guarentee
it.

It is often the default to create a new dataset to be of
type "short" (2 byte integers), along with a scaling factor.
This combination allows one to have numbers in basically
any range, but with only 15 bits of accuracy (plus the sign).

As an example, suppose you were to do a computation in 3dcalc
that produced a dataset with values between -10.0 and 8.4,
with the resulting dataset being of type short. Since the
maximum magnitude of the data is 10, your scale factor will
be 10.0/32768. This will allow you data values between -10
and 10, with a precision of 0.0003052.

The actual values in such a dataset would range between
-32768 and 26214 (corresponding to [-10,8]), and the scale
factor would be .0003052. A data value of 10000 would
correspond to the value 3.052 in afni.

So you are able to look at the data as integers. But in
afni and 3dcalc, etc, the data would be treated as floating
point numbers.

Basically, that is a way to half the size of the file, with
little loss of accuracy. It is uncommon for people to need
more than 5 digits of accuracy, and for those cases, the
true floating point dataset is available.

- rick