He creates his baseball predictions by matching current players with similar past players. Similarly, for each congressional district, he found another district that had already voted whose voters had similar education levels, income, racial mix, religious makeup, etc. Odds are, he figures, the votes in the two districts will be pretty close. He combines this demographic analysis with the polling data to get his best guess.
On May 6, his methods were put to the test. The latest polls said Clinton had drawn close to Obama in North Carolina. But Silver’s model didn’t buy it. In South Carolina and Virginia, Obama had done much better than the polls predicted. When the model matched each district in North Carolina with a similar one in South Carolina or Virginia, it calculated that Obama would clean up with a 17-point win.
Pollsters jeered. On Slate.com, Mickey Kaus said the prediction was made “by a blogger using a sophisticated model that ignores ... what's been happening in the campaign. Like Rev. Wright. I predict this person is wrong!”
But when the results were in, Obama won by 14 points. Silver beat every established pollster. And Poblano became a sensation, an instant election authority.
Here's a recently published interview with Silver; here's his Wikipedia page.