> -expr
> 'a*step(step(b)+step(c)+step(d)+step(e)+step(f)+step(g)+step(h))'
> where a=the f stat and b-h are the COEFS for each of the
> individual conditions that i would like to include.
Well, a>0 and all those steps make sure that what's to the right will be >0, so that's why you're only getting positive values. If you did something like:
a * ( (2*step(b)-1) * (2*step(c)-1) * ... (2*step(h)-1) )
I think you'll have what you were trying to get at -- the combined sign of all of the effects. Whether it makes sense statistically may be another story. Out of curiosity, if you just have these conditions, why not use the overall model's F? Also, if you have just these conditions and you want to see whether a region is active or "deactivated" -- relative to what?
Craig
