The Predicted Midterm Landslide
The “Seats in Trouble” forecasting model predicted in mid-August that Republicans would gain a landslide number of seats in the 2010 elections to the U.S. House of Representatives, and that this number would be sufficiently large to restore their majority control of the House, which was lost in the 2006 midterms.Footnote 1 Republicans were predicted to gain approximately 51 or 52 seats, about the magnitude of their 1994 midterm victory and the largest seat change since the Truman-Dewey election of 1948. As predicted, on Election Day, Republicans won a landslide number of seats, enough to give them a substantial House majority.
The “Seats in Trouble” forecasting model (really, two companion models) is a hybrid that draws on the different strengths of the in-depth district-by-district analysis of expert congressional election observers and the rigor and broader perspective of a statistical analysis of historical election data. The core of the model is a “seats in trouble” index calculated from the pre–Labor Day assessments of House races by the Cook Political Report. A seat is considered to be “in trouble” if it is not rated by the Cook Political Report as being solidly, likely, or leaning toward the party occupying the seat. The seats-in-trouble index is the difference between the number of Democratic and Republican seats in trouble. Many more Democratic than Republican seats were in trouble this year than in any previous election for which the index could be calculated, including 1994. In addition to the index, the models use a presidential approval measure and the number of seats that a party held after the previous election. The historical relationship between these predictors and actual seat change is estimated by a regression analysis over the 11 elections in which data are available to calculate the seats-in-trouble index.
The forecast of a 51- or 52-seat swing in favor of the Republicans was at the extreme end of the range of actual seat change over the last half-century of elections, as well as at the extreme end of predictions for 2010. In the early fall, only one other forecast (that of Bafumi, Erikson, and Wlezien) predicted that Republicans would gain more than 50 seats. Other forecasts were much closer to the 39- or 40-seat tipping point that was necessary for the Republicans to displace the Democrats as the majority party; still others suggested that the Democratic majority was not in serious jeopardy. More impressionistically, some pundits who offered forecasts were on the fence about Republicans retaking the House, and some hazarded a guess that Republican gains might be as great as the mid- to high 40s. In short, the seats-in-trouble forecast was not a common or safe prediction, but it turned out to be more accurate than other pre–Labor Day forecasts, equaled by only one other.
As accurate as the forecast was this year, there is still a significant error to examine. No one predicted larger Republican gains in August than 51 or 52 seats, but actual Republican gains amazingly exceeded this “out on a limb” forecast. Republicans gained 64 or 65 seats, with the votes in one race still being recounted at the time of this writing. Political observers generally considered the 54-seat swing in 1994 to be a political tsunami, but the 2010 results exceeded even that number by 10 or 11 seats. When the 112th Congress takes office in January 2011, there will be more Republican members of the House than in any Congress since 1946. While accurately predicting a Republican landslide, the forecast significantly underpredicted the extent of Democratic devastation by 12 to 14 seats (depending on the final district's recount). At the low end, this number is the size of the Missouri delegation to the House. At the high end, it is about the size of New Jersey's delegation to the House.
What might account for this forecast error? First, because of data limitations, the model is based on only 11 elections. This does not offer much leverage and may be the basis for some of the error. Second, the extent of seat change in 2010 was simply unprecedented in recent electoral history. Republican gains were “outside the box” of seat change examined in the model. Third, even if we had stronger models with better indicators over a longer series of elections, it may not be realistically possible to forecast House elections within a dozen seats or more—at least not wave elections with large changes. At this point, it appears that 35 races were decided within a margin of four percentage points of the vote, changed by a swing of two points or less. A few thousand votes here or there could have swung a large number of seats. One might expect that these races would evenly divide in the long run, but a forecast is not made in the long run. In this light, the results of the 2010 election were not too far from what the seats-in-trouble forecast expected.
