Political Weblog | 2006 Projections | 2006 Formula | 2006 Election Calendar | 2004 Results | Poll Vault | The Blogging Caesar Bio

 Election 2006 Projection Formulas


Senate and Gubernatorial Races

The formula I use for the Senate and gubernatorial races is comprised of 4 possible components.  The driving idea behind my formula is that an average of polls from different sources is an accurate measure of the current state of a political race.  Therefore, this poll composite claims the bulk of consideration in my formula.  Sometimes, however, polls can miss certain underlying factors that may render them less accurate than they otherwise might be.  I've tried to capture and quantify the impact of such factors by including additional metrics in the formula.

One of them is the job approval of the incumbent.  Obviously, this factor is more relevant when the incumbent is seeking re-election.  But it also impacts the incumbent party's chances in open seat races.  My formula gives a 15% weight to this component in races where the incumbent is running and 5% in open seat races.

The next component in my formula measures the partisan tendencies of the state's electorate.  If a state is historically Democratic or Republican, more voters there are likely to break toward their party of choice in the end.  We see this phenomenon in states such as New Jersey on the Democratic side and some of the southern states for the GOP.  In these cases, polls notoriously inflate the minority party's standing.  My formula tries to mute that effect.  State partisanship is calculated by subtracting Kerry's vote total in the state in 2004 from Bush's total.

Finally, there are rare cases when extenuating circumstances necessitate tweaking the numbers just a bit.  For instance, a possible indictment for a candidate or a party whose public image has been in shambles are cases where polls and approval might not completely reflect the actual condition of a race.  As a specific example, the GOP in Ohio has come upon hard times of late, and the effects of their predicament probably outpace the apparent downturn in the polls.  Because of this, you'll see an adjustment of 0.5% in favor of the Democrat in the Senate race here.  (Because GOP Governor Taft's approval numbers are so abysmal, the same adjustment is not employed in the Governor's race.)  As this component is only a method a tweaking the result, it shall never be larger than one percentage point.

So, with that explanation, I present the 2006 Election Projection Formula for Senate and gubernatorial races.

For incumbent seeking re-election:

Poll Avg. * 0.8 + Inc. Net Approval * 0.15 + State Partisanship * 0.05 +/- Adjustment  =  Projection

For open seats:

Poll Avg. * 0.9 + Inc. Net Approval * 0.05 + State Partisanship * 0.05 +/- Adjustment  =  Projection

Here's an example of the formula in action - these numbers are the actual numbers for the New Jersey Senate race at the time this formula was published (07/15/06).



  • Rasmussen:  Menendez 46%, Kean 40%
  • Strategic Vision:  Menendez 43%, Kean 37%
  • Rutgers-Eagleton:  Menendez 42%, Kean 38%
  • Quinnipiac:  Menendez 43%, Kean 36%

    Menendez's Net Approval:  -1%

    New Jersey Partisanship:  6.7% Democrat

    Other Issues: 0.5% Democrat - this is due to the fact that the GOP is famous for polling considerably better here than they actually perform.


    Positive numbers favor the incumbent party

    Poll Avg. * 0.8 + Inc. Net Approval * 0.15 + State Partisanship * 0.05 +/- Adjustment  =  Projection

    5.8 * 0.8  +  (-1) * 0.15  +  6.7 * 0.05  +  0.5  =   5.3


    Menendez is projected to defeat Kean by 5.3%

    Contested House Races

    The formula for this year's House races is substantially unlike the formula I described above.  There are four main differences between this formula and the one used for Senate and gubernatorial races.  First, since job approval numbers for House members are extremely difficult to track, there is not a job approval component to the formula.

    Second, published polls for district races are also much more difficult to find than for Senate or gubernatorial races, so I needed to find another metric which would be reliable and consistent throughout the election season.  I decided to use an average of the predictions of 3 well-known pundits and The Blogging Caesar.  In addition to my own, I'm going to be using the predictions of Charlie Cook, Larry Sabato, and the Congressional Quarterly.  Since their predictions are qualitative (leans, toss-up, solid, etc), I will assign a margin of victory to their predictions as follows:

      Toss-up/No Clear Favorite Lean Likely Solid/Safe
      0% 4% 8% 16%
    An additional designation will be made for Larry Sabato's predictions.  He ranks the top thirty most competitive seats and calls them his "Dirty Thirty".  Some of these are toss-up and some are considered to be leaning to one party or the other.  Beyond that, he has a "Watch List" of 20 more which are the next tier of competitive races.  These are also rated as "Lean".  For the purposes of this formula, "Leans" in the Dirty 30 are assigned a 4% margin while races in the Watch List receive a 6% margin.

    Third, when polling data can be obtained, their weight be will based on whether one poll or multiple polls are available.  If one poll is available, it will get a 25% chunk of the projection.  If a poll has been published by more than one polling firm, the average will be taken, and that result will be 50% of the projection.

    Finally, district partisanship is used instead of state-wide partisanship.  And this measurement is calculated differently, too.  Since the advantage of incumbency is quite large in House races, I decided to use the incubment's performance in the previous election as part of the partisanship metric.  Here's how I do it.  If the incumbent is running for re-election, his or her margin of victory in 2004 is averaged with the outcome of the 2004 presidential race in the district.  The result becomes the district's partisanship.  If the incumbent is not seeking re-election, only the presidential result is used.

    Here then is the 2006 Election Projection Formula for contested House races.

    For races with no published polls

    Pundit Avg. * 0.95 + Dist. Partisanship * 0.05 +/- Adjustment  =  Projection

    For races with one published poll

    Pundit Avg. * 0.70 + Poll * 0.25 + Dist. Partisanship * 0.05 +/- Adjustment  =  Projection

    For races with two or more published polls

    Pundit Avg. * 0.45 + Poll Avg. * 0.50 + Dist. Partisanship * 0.05 +/- Adjustment  =  Projection

    Here's an example of this formula in action - these numbers are the actual numbers for the race in New Mexico CD-1 at the time this formula was published (07/15/06).


    Pundit predictions:

  • The Blogging Caesar:  Madrid by 3%
  • Charlie Cook:  Toss-up
  • Larry Sabato:  Toss-up
  • Congressional Quarterly:  Wilson by 4% (lean GOP)


  • Lake and Associates:  Wilson 44%, Madrid 43%

    New Mexico CD-1 Partisanship:  2.5% GOP (avg. of Kerry's 3-pt and Wilson's 8-pt win here in 2004)

    Other Issues: 0.0%


    Positive numbers favor the incumbent party

    Pundit Avg. * 0.70 + Poll * 0.25 + Dist. Partisanship * 0.05 +/- Adjustment  =  Projection

    0.3 * 0.7  +  1.0 * 0.25  +  2.5 * 0.05  +  0.0  =   0.6


    Wilson is projected to defeat Madrid by 0.6%