Hello Carl,
The only time I have seen F stat signed is for display purposes, it is of course meaningless as a statistic. Some people, who shall remain nameless, multiply the sign of a regressor's coefficient by its F value as a simple way to utilize the entire range of a two tone colormap.
That is done by using the following equation assuming C is the sub-brick with the coefficient and F is the sub-brick with the statistic.
-expr ' F * (step(C) - step(-C)) '
step(C) = 1 when C > 0
= 0 when C < 0
step(-C) = 0 when C > 0
= 1 when C < 0
so
( step(C) - step(-C) ) = 1 when C > 0
= -1 when C < 0
I do not understand why you are trying to sign the Full F stat. Unlike the partial F, this one has no coefficient that goes with it because it is testing whether any of the coefficients was significantly > 0. What are you defining as negative activation ? What if one stimulus caused a negative response while another cause a positive one ? If you insist on signing the F stat, perhaps you should only sign the partial F stats.
cheers.
-z
