[SomeCallMeTim]

There are more ways than one to skew a poll. NPR had +4%D sample, ie oversampling Democrats, but not as much as some other polls. But the poll also oversampled women:

54% women/46% men. That will skew the results.

Correction: In 2008, the breakdown of actual voters was: 53.7% Women
46.3% Men

It's a surprisingly hard statistice to come up with. But, that's what I calculated, using actual voter numbers.

[ScaniaBoy]

I knew that women voted slightly more often than men, but a 54/46 difference! I’m really taken aback. What did it look like in 2010?

(It does explain why the result turned out like it did in 2008!!)