Getting a handle on which way the races are trending is at once easier and harder than it used to be. Easier, because in this wired world everybody is offering up predictions; harder, because many aren't reliable--they're uniformed, overtly partisan, or fly-by-night hucksters. There's no Good Housekeeping Seal of Approval.
But James E. Campbell, a political scientist at SUNY Buffalo, has done something very useful and informative. He went and checked the House race predictions of one venerable prognosticator, the Cook Political Report, going back to 1984 and measured them against the election outcomes to see how good they were. It turns out they were very good. Across 11 elections (1984-2008) the Cook Report correctly called races rated "solid, likely or leaning to the Republicans" a whopping 97.5 percent of the time. On the Democratic side, the Cook Report was right 97.8 percent of the time.
Now you might be thinking to yourself, "Anybody can make a good prediction a week from Election Day." And that's what I think is most noteworthy about Campbell's study: he only measured the ratings issued before Labor Day, which is to say, all of these calls were made a good two months before the election.