To all users of the AFNI Permutation Test plug-in:

I've just released version 1.2 of the Permutation Test plug-in. THIS IS AN
IMPORTANT BUGFIX RELEASE; EVERYONE WHO USES THE PERMUTATION TEST SHOULD INSTALL
IT.

In very rare cases (which is why it wasn't discovered till now), this bug in
versions 1.0 and 1.1 of the plug-in can cause AFNI to enter an infinite loop
in Phase 3 of the permutation test algorithm. Furthermore, the bug can add a
very small (which is another reason it wasn't discovered till now) amount of
noise to the computed tail probabilities, errors on the order of 0.0001. (This
is in comparison to typical alpha levels of 0.05 -- thus it should not be
necessary for you to re-run previous computations.)

To check what version of the Permutation Test plug-in you're running, start the
plug-in and then click on its `Help' button; towards the bottom of the help
window you'll see a revision history. You need to be running revision 1.2.

Version 1.2 of the AFNI Permutation Test plug-in is available as source code
from "[www.mattababy.org];,
an archive which also includes an updated version of the file README.permtest.
I've also temporarily placed a pre-compiled binary for Linux at
"[www.mit.edu];.

For those of you keeping score at home: The bug was in the function
dtree_unlink_node(), which is used in the process of removing a coordinate
triple and its associated correlation value from the computed distribution,
after these coordinates have been marked as activated. When the relevant node
is deleted from the tree structure that holds the computed distribution, the
rightmost node of its left subtree is promoted to fill the resulting vacancy.
In the case in which the left subtree of this rightmost node is non-null, the
bug caused this subtree's parent pointer to be given an incorrect value, one
that was further towards the root of the tree than the node's actual parent
and was thus inconsistent with the value of the associated child pointer.

Greg Allen of the University of Texas Southwestern Medical Center kindly
submitted the bug report that uncovered this problem. That's some good
bug-huntin', Tex.

Big plans are afoot for the forthcoming release 2.0, which you can expect in
late spring or early summer. As I mentioned a long time ago on the AFNI
bulletin board, it's on my list of things to do to add Ed Bullmore's
wavelet-based pre-whitening procedure to the permutation test. That task has
lately been bumped up on my list of things to do, since I'm about to be going
to a higher field and lower TR's where temporal autocorrelation starts to
matter. Anyone who'd be interested in running pre-release versions of this
code should contact me. (No guarantees on when the pre-release version(s) will
be available, though.)

As always, I'm interested in hearing from anyone who's making use of the
permutation test.