Senate and House Ballot Forecast methodology

by: Chris Bowers

Wed Jan 20, 2010 at 13:40


  1. The forecast is entirely based on polling.

  2. For now, use the simple mean of all polls where the majority of interviews were conducted over the last 30 days.

  3. During the final six weeks of an election, use the simple mean from the last 15 days.

  4. When available, at least two polls are used for every matchup, even if the majority of their interviews were conducted more than 30 days ago.

  5. Do not include Zogby Interactive polls and Columbus Dispatch polls, due to their horrendous past performance and questionable methodologies.

  6. Do not include Strategic Vision polls, as it is starting to seem likely those are not real polls. Same goes for Research 2000.

  7. Include partisan and campaign-funded polls, as long as the full results are released to the public. (The more polls, the better).

  8. Further, if there is more than one poll from a single organization, include all of them. (The more polls, the better).

  9. The win % represents the chance for a Democratic victory given the current polling averages. It is based on my research into the 144 closest statewide campaigns from 2004-2010. For example, across those 144 campaigns, there is a 50% chance for the final polling average to differ from the final result by more than 1.76%. Thus, a candidate ahead by 1.76% is given a 75% chance of winning, since the polling average has an equal chance of being skewed in favor of either major-party candidate.

  10. In campaigns where the general election matchup has not been decided, the win percentage used in the overall forecast features the two candidates who currently lead their respective primaries.

  11. Campaigns where one party is ahead by 6.0 or more are considered "solid" (95% chance for leading party to win). Campaigns between 1.8 and 5.9 are considered "leans" (75-95% chance for leading party to win). Campaigns within 1.7 or less are considered "toss-ups." These categories are represented by the colors in the charts, and are included to make the chart more readable.

  12. Candidates that have not officially entered the campaign are not included in the forecast.
The research and rationale behind this methodology can be found here.

Across the 52 closest general elections for Senate, Governor, and President (both swing state and national) from 2008-2010, here is how this methodology compares to and's:

Error rates, final predicted margin to final vote margin, 52 campaigns, 2008-2010
Pollster 538 Simple 15-day mean
Mean error 2.79 2.83 2.61
Median error 2.14 2.16 1.77
It may be comparatively simple, but it seems to be effective.  You can download all of the data for this comparison here.

Notes: These 52 campaigns were chosen because they were the closest 52 campaigns for which both and published final estimates.  

The 2008 Nebraska Senate race was not included because there were not enough polls for's Loess regression methodology to be used.

Chris Bowers :: Senate and House Ballot Forecast methodology

Tags: , , (All Tags)
Print Friendly View Send As Email


Open Left Campaigns



Advanced Search

Powered by: SoapBlox