Running the numbers: Obama/McCain polling and changes in their support

At the time of writing, The American Research Group has provided some statistics on the American Presidential election. What grabbed my attention was the detailed breakdown of some of the numbers. Now, it's possible to tie yourself into all sorts of ridiculous knots with this kind of demographic detail with questions about how Candidate A is faring against Candidate B amongst female immigrant blue-collar dog owners. Nevertheless, this is eye-catching:

National
Likely voters	Aug 30- Sep 1	Sep 13-15	Change (pts)
McCain:
Democrats	5%	19%	+14
Republicans	83%	84%	+1
Independents	43%	43%	0
Obama:
Democrats	88%	75%	-13
Republicans	11%	9%	-2
Independents	45%	49%	+4

This is the breakdown of support for each candidate among Democrats, Republicans and Independents, respectively. Note the striking swing among Democrats away from Obama and towards McCain. There has been a slight swing away by Republicans as well, but not a huge amount, and Obama's gained some Independent support (although McCain hasn't lost any). But it's that 2-digit swing towards McCain from people who identify as members of Obama's own party where Obama's hurting most, to my amateur eye.

The cause? Difficult to say, but it's tempting to view it at as a gender issue. After all, the same period shows Obama losing 5% of women and McCain gaining 7%. But that's ALL women, not just Democrats. Nor is it right to assume that the Democratic support Obama's lost is mostly female, although it very well could be.

Both surveys were taken after Obama had officially accepted the nomination instead of Hillary Clinton, so that alone doesn't explain it. The first survey was taken after Sarah Palin was announced as VP for McCain, but before she gave her convention speech. That's a possible area for analysis. But given the two week gap - a very eventful two weeks - between surveys, it's difficult to pinpoint the exact issue. More data is required. Unfortunately, I suspect that the required data is more detailed than most survey companies are willing to put up for free on their websites.