Hello Jennifer:
Yes, it's wrong to arbitrarily remove the "negative activations before entering
each subject's functional map into the ANOVA". Just because "people rarely
speak about their negative activations" doesn't mean that they're not real.
The ANOVA analysis (as well as the t-test) make certain assumptions about the
underlying populations when the null hypothesis is true; such as normality,
constant variance, and independence of observations. If you suspect that
"negative activations" may represent outliers in the data, then you may
prefer to use a nonparametric statistical approach, which is more robust
to outliers in the data. Such tests are provided by programs 3dMannWhitney,
3dWilcoxon, 3dKruskalWallis, and 3dFriedman, which are described in the
documentation contained in file Nonparametric.ps.
Doug Ward