******** Changes to the 3dANOVA Suite ********

We (the SSCC group at NIMH/NIH/DHHS) have recently made modifications to
3dANOVA, 3dANOVA2, and 3dANOVA3 (1-way, 2-way, and 3-way Analysis of Variance
programs for group analysis).  Some of these changes might considerably
affect the outcome of simple effect (amean, bmean) tests, and general linear
tests whose coefficients do not add up to 0.

This notice serves as an alert; much more detailed information, and
examples of pre- and post-modifications can be found at this Web page:
   http://afni.nimh.nih.gov/sscc/gangc/ANOVA_Mod.html
 
The first and potentially largest effect of the modifications involves
the calculation of the statistic for simple effect (amean, bmean) tests,
but NOT that of zero sum general linear test (contrast, -acontr, -bcontr)
or main effect tests (-fa, -fb).  This change does NOT affect 3dANOVA, just
3dANOVA2 and 3dANOVA3.
 
For the affected tests, the variance term (denominator) for the t statistic
is from the sums of squares of the interaction.  Such an estimate would be
correct if inter-subject variance was negligible.  However, this is generally
not the case in FMRI.  As a result, the t values for simple effect tests and
general linear tests whose coefficients do not add up to 0 were artificially
elevated, resulting in more widespread activation.

In the new version, a more appropriate denominator is used, which includes
inter-subject variance.  This change brings simple effect testing in line
with what one observes with a simple t-test (e.g., with program 3dttest).
Comparing the new to the old simple effect maps, you will notice reduced
areas of activation, but little change in the location of the activation
peaks.  See
   http://afni.nimh.nih.gov/sscc/gangc/ANOVA_Mod.html
for a visualization of the differences in one example.
 
A second effect modification involves making tests less vulnerable to
non-sphericity (e.g., correlations in the noise terms).  All tests are
affected by these changes; however, the differences in the results are
usually very subtle and hard to notice in images.  See
  http://afni.nimh.nih.gov/sscc/gangc/Contr.html
for the mathematical details of this change.
 
The newest versions of the programs allow you to force the programs to run
in the old mode.  This will allow you to reproduce results under the old
assumptions and compare them to the current ones.

We cannot tell you if YOUR results will change significantly between the old
and new versions of the calculations; the only way to find that out is to
run the ANOVA computations both ways yourself on your actual group datasets.

Summary:
 - If you were using only contrast tests whose coefficients add up to zero,
   then results calculated the old and new ways will probably differ only in
   very minor details.
 - If you were using tests whose coefficients did NOT add up to zero, then
   it is possible (but not certain) that the areas of activation calculated
   the old way are significantly bigger than the areas calculated the new way.
   The locations of the peak activations probably will not be very different.

In addition, some new options have been added, to make it easier to calculate
various contrasts.  These are described at the SSCC Blog:
  http://afni.nimh.nih.gov/sscc/log/blogentry.2005-12-29.4898565589
and in the '-help' output of the programs.
------------------------------------------------------------------------------
 Questions about these changes should be addressed on the AFNI Message Board!
------------------------------------------------------------------------------
----------------- Szczesliwego Nowego Roku - Happy New Year! -----------------
------------------------------------------------------------------------------