O+2 R-2. Think vector notation. Romney needs to gain 2 points to tie Obama, since a gain for Romney will be the same as a loss for Obama.
And yet, you won't accept an appeal to definition when it concerns statistical significance.
So you're willing to be 20:1 odds that the next flip is heads?
This is why there's no correlation. It is, as I stated before, a mathematical equivalency. Not arbitrary whatsoever and is fundamentally connected with sample size.
Weak correlation is when you have multiple studies, some with statistical significance, and some without.
Glad I could help you pass your stats class!