Data for the House Elections Predictor
This page describes the data used in my projections. If you wish to read about how the projections work, and how accurate they are, I've written a page about the projections.
Presidential Approval Data
The data analysis underlying my projections is based on Gallup's 
presidential approval ratings. (Though other organizations also collect presidential approval data, they use different methodologies, so you should use Gallup's data for the best predictions.) The Roper Center maintains an archive of approval ratings that you may find interesting.
In elections from 1946 to 2004, presidents entered November with approval ratings anywhere between 32 and 74, though most were between 42 and 61. So far, Bush's approval ratings have been low. Though Gallup collects data more than once a month, I'll put monthly ratings here (if I remember):
Economic Growth Data
The model also considers how much real income per capita has changed over the past 12 months. In elections from 1946 to 2004, this has been between -2.6 and +6.0, but usually between 0 and 3.3. For current data, visit check the Bureau of Economic Analysis . Follow the link, scroll down to Table 10, and look at the first row of data in the table. Note that BEA has moved where it publishes this statistic, and I haven't found it yet. If you find where it is on their site, please let me know. Unfortunately, these figures are released with a one-month lag (sometimes more), which limits our ability to make good projections. For your convenience, here are some of their recent figures. Note that the BEA is likely to revise these data over the next few months:
Exposure
If the president's party won more seats than usual in the last election, that implies that the president's party has more opposition-leaning districts to defend than usual as well. As such, the model accounts for how the incumbent party's current seat share compares with its average share over the previous eight election cycles. The following graph gives an idea of how much partisan seat shares can fluctuate. (Click the image to enlarge it.) Note that the graph covers every House election since America's founding, although this model uses only post-WWII elections. Since 1946, exposure has ranged from -19.4% to 11.1%.
