These figures caught my eye from TPM’s summary of the Insider Advantage Pennsylvania poll (which showed Clinton with a 2-point lead):
From the internals: Clinton leads only 49%-40% among whites, while Obama is ahead 56%-29% with black voters. Clinton is ahead 49%-38% with women, while Obama is ahead 47%-41% with men.
Obama’s “small” 56 to 29 lead among black voters surprised me. In Ohio, which people seem to think is the closest there is to a similar state, Obama won 88 percent of the black vote, according to exit polls. In New Jersey, another neighboring state that Obama lost to Clinton, he won about 84% among black voters. I don’t remember seeing any state where he won under 70% of the black vote. The Quinnipiac Pennsylvania poll, which was released 2 days ago and showed Clinton with a 9-point lead, had Obama with a 73 to 11 lead among likely black voters.
So I don’t really know what to make of those Insider Advantage numbers, although I don’t pretend to understand the subtleties of administering or interpreting polls. I also notice that in Quinnipiac and Insider Advantage both, the “undecided” (or “won’t tell a pollster”) vote is a bit higher among black voters, so perhaps there’s some kind of “reverse Bradley effect” going on where black voters tend to be less likely to tell pollsters who they plan to vote for, and when they do tell, they are more likely to feed the pollster misinformation. But that still wouldn’t explain the strange discrepancy where Obama’s lead among black voters is actually lower in the poll where Clinton’s overall lead is smaller. Maybe it’s just an example of a small sample size producing an unreliable result?