RAND has an interesting tracking poll (American Life Panel) which interviews the same 3500 people (1/7 each day) repeatedly over the course of several months.
This obviously can introduce some biases as those people become more aware of the election than others might be, but also removes most random sampling noise: shifts that you see are due to real people changing their minds.
As I diaried earlier, they have recently begun to put the now-cast probability for Obama at (wait for it) 100%. That number is not presented directly by them, but can be computed by inserting their results (49.84 Obama, 43.93 Romney) into this Ballot Lead Calculator.
In addition to the overall results for a few indicators, they also publish a breakdown each day for one of the indicators along various demographic lines. Today they show an extremely interesting breakdown of election forecast by sex. This shows that virtually all of the recent swing towards Obama comes from males. A week ago, they showed Romney with a statistically significant lead of 6.5% among males. Today they are at near parity, with Romney just 0.72% ahead among males. Meanwhile Obama has held a steady and commanding lead of about 12% among women.
This follows a similar lop-sided shift among just the women a couple of weeks earlier. Could it be that women are more perceptive? Or were there two events that influenced each group separately? And has anyone noticed anything similar to this in past elections?
10:49 AM PT: Here's a snapshot of the current forecast by sex, as these graphs will be replaced tomorrow by others at the RAND site. (Thanks to assaf for guidance.)
RAND corporation American Life Panel tracking poll, morning of Sep. 22 2012, broken down by gender.