It's a 600-person poll, and Detroit city is less than 7% of Michigan. 7% of a 600-person poll is a sample of 42. A sample of 42 on a population size of 871,121 gives a margin of error of greater than +/-15%. So...not much to see there.
Ya, and McCain got the entire 15% plus side of the bell curve tail into his courner in Detroit and then some. LOL.
I'm not sure how EPIC does that screening, whether it's touch-numbers or based on phone exchanges. If it's the former, it's quite plausible that there's some noise involved. If it's the latter...who knows? But lately I get the impression that the "1-in-20" rule of samples is trending a little north of the 1:19 ratio.
Especially with what's a rather bad pollster.