Democrats are currently leading in national "generic ballot" polls that ask people which party they prefer for House races (without naming candidates). That fact has led some forecasters, such as Princeton's Sam Wang, to conclude based on past elections that Democrats are favored to retake the House. Wang puts the odds of that at 74 percent.
But Kevin Drum isn't convinced. He argues that the generic ballot congressional polls tend to overstate Democratic support in the national popular vote by about 4 to 5 percentage points. And, since the national popular vote does a much better job of predicting House elections, this suggests that Wang is far too optimistic about the Democrats' chances.
So is this true? Does the generic ballot overstate Democratic support? It's a key question in trying to figure out which party will control the House.
In the past, the generic ballot has overstated Democratic support, as polls from 1992, 1988 and 1980 demonstrate. But the polls' predictive power has increased in recent years. The best work on this has been done by Dartmouth's Joseph Bafumi, Columbia's Robert Erikson and Temple's Christopher Wlezien. Their model uses national generic ballot polls, as well as which party holds the presidency, to estimate how many seats each party will pick up.
And this model performs quite well. The researchers predicted that Democrats would gain 32 seats in 2006 — the actual number was 30. The model predicted Democrats would lose 50 seats in 2010 (Democrats actually lost 61). It's not perfect, but it fits past elections, even if you use polls from a year before the election:
What does this mean for 2012? Using Bafumi, Erikson and Wlezien's model, and the last month's average result of a 2.2 percent Democratic victory margin, you get a Democratic victory margin of 0.688 this year. That translates to a razor-thin one-seat margin, if you only look at how vote margins have translated into seat margins in elections from 1996 onward, or a 22-seat margin if you look from at all elections from 1946 to 2010 (thanks to Sam Wang to passing along this data).
And with a typical error of 18 seats in either direction, that means Democrats have hardly got it in the bag. Moreover, Bafumi, Erikson and Wlezien's regression was for October data, and may be less reliable with data this far out. But this does suggest the Democrats have a good shot at retaking the House.