I don't trust those SurveyUSA crosstabs. They are usually horrible for any poll, but they seem to be especially the case with SurveyUSA.
SUSA always gives very full crosstabs, which look pretty, but are often crazy due to small sample size. I seem to recall in 2008 they had polls with Obama winning places like WI and IA decisively, but losing with black voters in those states. Then you looked, and saw they'd (quite properly) interviewed about a dozen AfAm voters, which makes those results meaningless.
If nationwide, Obama was winning 11 out of 12 blacks, and in some state McCain won 7 out of 12 blacks, that's a statistically significant difference.
