About the House Elections Predictor

This page answers general questions about how the projections are made and their accuracy. To learn about the data used in the projections, visit the data page.

Disclaimers

Projections are based on an historical analysis in Gary Jacobson's The Politics of Congressional Elections, 6th ed (2004). Projections assume that the values you enter are measured just before election day. These projections have a wide margin of error of several seats. Results not guaranteed.

If the default values seem out of date, to report a problem, or if you have a question, please tell me about it.

Does this work? Why?

Many visitors wonder why Congressional swings can be predicted using aggregate data. This question is reasonable; after all, Congressional elections are decided entirely by each district's voters, and individual voters rarely mention presidential performance or national economic performance when explaining their preference in Congressional elections. Yet Edward Tufte demonstrated that presidential approval and economic performance had strong, persistent correlations with aggregate swings (see cite 1 and cite 2 ). Later scholars identified the importance of "exposure"--in other words, given the historical representation of the president's party in Congress, is it currently over- or under-represented (see cite )? (For the statisticians out there, this "exposure" concept controls for the problem of regression to the mean.)

But if national factors are irrelevant to a single district's campaign, why is there such a persistent correlation between national factors and aggregate swings? It turns out that national factors affect each party's ability to recruit quality candidates and to raise campaign contributions (see cite ). In a House race, a "quality" candidate is generally somebody who currently holds some other elective office--mayors, state legislators, and so on. These experienced candidates are less willing to run for the House--and risk losing their current office without gaining a House seat--when their party seems to be in a lull. Similarly, campaign donors are less willing to contribute money to a candidate with little chance of winning. This, then, is the solution to the puzzle. Because candidates and donors respond to national conditions (primarily presidential popularity and economic performance), they drive the national swings.

For an up-to-date discussion of these issues, read the most recent edition of Jacobson's The Politics of Congressional Elections.

Are these predictions accurate?

Many visitors are surprised to see that the president's party is predicted to suffer a large loss in midterm elections. They are equally surprised to learn that large losses are the norm. From 1946 until 1994, the president's party lost House seats in every postwar midterm election, with an average loss of twenty-four seats. Actual swings from 1946 to 2004 vary from a loss of 56 seats to a gain of 8 seats.

That being said, you should take this site's projections with a grain of salt. (Here comes the statistical mumbo jumbo.) It is based on an historical analysis of the House elections from 1946 to 2004. The statisticians out there will want to know that the equation explains 72 percent of the variance (adjusted R squared = 0.69) in House election swings, and that all independent variables are statistically significant at p<0.001 (two-tailed). Still, the margin of error around these predictions is large. The margin varies depending on the specific numbers that you input, but for most predictions using inputs actually observed in the past the margin of error (for a 95% confidence interval) will be in the neighborhood of +/- 20 to 50 seats. This margin of error reflects the confidence interval for the average predicted seat swing given the inputted values of X. Note that a confidence interval around the specific forecast for a given year would be much wider, and would almost certainly include zero. (I'm working to make the site return a confidence interval with each point prediction, but doing so may take awhile.)