Bafumi/Erikson/Wlezien predict a 50-seat loss for Democrats in November

Posted on by

They write:

How many House seats will the Republicans gain in 2010? . . . Our methodology replicates that for our ultimately successful forecast of the 2006 midterm. Two weeks before Election Day in 2006, we posted a prediction that the Democrats would gain 32 seats and recapture the House majority. The Democrats gained 30 seats in 2006. Our current forecast for 2010 shows that the Republicans are likely to regain the House majority. . . . the most likely scenario is a Republican majority in the neighborhood of 229 seats versus 206 for the Democrats for a 50-seat loss for the Democrats.

How do they do it? First, they predict the national two-party vote using the generic polls (asking voters which party they plan to vote for in the November congressional elections). Then they apply the national vote swing on a district-by-district level to predict the outcome in each district. They account for uncertainty in their predictions (I assume by using a model similar to what Gary King and I did in 1994), which induces a probabilistic forecast of the number of districts won by each party.

Regular readers will know that this is not news. Way back in September, 2009–over eleven months ago–we used an earlier version of the Bafumi/Erikson/Wlezien model to predict a Republican House takeover in 2010. As I wrote several months ago, it feels good for once to be ahead of the story.

5 thoughts on “Bafumi/Erikson/Wlezien predict a 50-seat loss for Democrats in November”

  1. An accurate prediction two weeks before election day (when many vote have already been cast in some states) is hardly impressive. The interesting unpredictable stuff often happens after the primaries and before the election in the couple months of general election campaigning.

  2. I told you you were wrong then and you are still wrong now. The reason is simple. The polls today do not mean what they used to mean. When the Dems lose 20-25 seats maybe you will listen. Of course, if I am wrong, i will bow to the wisdom of poli sci.

    Here is why the polls are wrong. They still track closely to economic performance and presidential approval (generally the same thing), but actual elections do not. I estimate that the impact of those factors is less than half what it used to be. The midterm elections now change as much due to voter turnover (new voters coming in, old voters dying) as economic performance. Blacks, Hispanics, and all voters born since 1978 vote (for Dems mostly) without regard to economics. In 2006, the demographics and the economic fundamentals moved in the same direction, so the old formula worked.

  3. Ohwilleke: Election day is more than two weeks from now, and it's certainly more than two weeks from September, 2009, when I first wrote about this.

    Ockham: Maybe so.

  4. It would be interesting to see how sensitive the prediction is the the generic ballot number, which is the only input number in the model that is visible. How much would a one or two percentage point shift in that number matter? How much would an outlier poll result blow the numbers?

    In other words, how much does this prediction deserve to be taken seriously?

    What has distinguished off the mark predictions in past elections? The conventional wisdom is that a lot of primaries have been going to more politically extreme, rather than less politically extreme candidates, and that this hurts insurgents. The model is too simple to consider this, but an analysis of the error patterns in past elections would shed insight into whether there is any merit in this analysis.

Comments are closed.