Rassmusen has established a pretty strict determination of who qualifies as a likely voter. The problem with that model is it is static. The actual voters in this election will not be the same as those Rassmusen is using. Other polls are attempting to catch changes in who is likely to vote and they seem to be determining that more GOP s will turn out and less rats will turn out. That seems plausible at this time.
The problem with establishing a too strong test of who a likely voter is, you can end up with samples that contain very strong representation from either end of the spectrum and none from the middle. Imagine running a poll that was comprised of equal parts of Freepers and DU'ers each day. That sample would not fluctuate at all.
These polls do make me believe that the entire electorate shifts back and forth on a daily basis... not possible
Six weeks will not come soon enough.