This page describes the variables used in the model, the algorithm used for four different types of races, and how we measure the accuracy of a projected party and winning percentage.  This information is common to the methods used for 1996-2000 as well as for 2002.

For the 2002 projections, however, we made two modifications to account for changes in districts due to redistricting and the lack of availability of data about the partisan nature of 60 of the 435 new districts.

First, we adjusted incumbents' prior winning percentages according to the change in partisanship of the 2000 district compared to the 2002 district.  For example, if a candidate received 58% in a district that became 5% stronger for the incumbent party, we would treat the prior race as a 63%.  If the district became 5% less favorable to the incumbent, we would treat it as a 53%.

Second, in districts that lacked 2002 partisanship data based on the Gore's percentage in the new district, we cautiously estimated a partisanship by subtracting 3% from the partisanship of corresponding 2000 district.

Third, because our knowledge of the 2002 districts is limited, we increased the parameter, "Decrease for long term overachievers," from 42% for 1996-2000 to 60%.  This means we make slightly more cautious projections for 2002.

In all other respects, the model treats races from 1996-2000 identically to 2002 races.


This section describes each variable used in the model, along with the default value and the category of seats it applies to.  The variables typically refer to an incumbent’s weakest performance out of the last two elections.

Variables related to projections

Buffer for open seats (11%).  This variable is amount by which partisanship is reduced to make an open seat projection.  For example, a 70% district is reduced to a 59% projection.

Increase for underachievers (33%).  If a freshman runs behind her partisanship in her first election, her projection is raised 33% of the way up from her election result to the partisanship.  For example, a freshman elected with 55% in 61% district is projected to win with 57% in her second election.

Decrease for overachievers (67%).  If a freshman or a two-term member runs ahead of their district partisanship, we reduce their projection 67% of the way back to the partisanship.  For example, a candidate who wins with 61% in a 55% district is projected to win with 57%.

Reduction for underachievers (1%).  If a two-term or three-or-more term incumbent’s worst performance is below the district partisanship, the projection is 1% less than their worst performance.

Previous race uncontested (35%).  If a two-termer’s second election was uncontested, the incumbent is treated like a freshman.  This variable represents the minimum winning margin by which a race is considered uncontested race.

Decrease for long-term overachievers (60%).  If a three-term incumbent’s worst performance is better than the partisanship of the district, the projection is reduced by 60% of the difference of partisanship and worst performance.

Adjustment to better 2nd election (33%).  If a two-termer’s second election was better than their first, then use this variable to interpolate from the stronger performance to the weaker performance to establish the “weakest performance” number.  The purpose of this is to give more weight to the 2nd election.
Variable related to national two-party vote

Dem 2-party share (50%).  Our model makes projections based on an assumption that the national two-party vote is evenly split between Democrats and Republicans.  To examine the projections in the event the two-party vote is not evenly split, you can enter a different number in this cell.

Variables related to categories

No proj win  (3%).  If a projected result is between 47% and 53%, it is considered “no projected win.”

Win (5%).  If a projected win is between 53% and 55%, it is considered a “win” projection.

Comfortable (10%).  If a projected results is between 55% and 60%, it is considered “comfortable.”

Landslide (>10%).  This is not a separate variable, as it encompasses all projections greater than “comfortable.”
Algorithms for 4 different types of races

I.  Open seats

Use partisanship to project outcome, but then subtract 11% off projected margin.

Ex:  a 62% partisanship is a 24% margin, minus 11% = 13% margin or 56.5% (comfortable)

There are 3 separate cases for incumbents.  For 2002 projections, we begin by adjusting past performance to reflect change in district partisanship from 2000 to 2002.

Ex:  A 53.5% win in a district that changed from 52.1% to 54.3% picks up 2.2% and is treated as a 55.7% result.

II.  Freshman

If past performance was below partisanship, the projection equals the past performance + 1/3 the difference with partisanship. (Projection stronger than past performance.)

Ex:  Past performance 42.5%, partisanship 48.5%, projection = 44.5%

If past performance was better than partisanship, the projection equals past performance – 2/3 of the difference with partisanship.  (Projection weaker than past performance.)

Ex:  Past performance 48.5%, partisanship 42.5%, projection = 44.5%

III. Two-termers

If 2nd election was uncontested, treat first election as like a freshman.

If 2nd election contested, then consider weakest of past 2 performances.

If the 2nd election was stronger than the first, adjust the past performance by interpolating from the stronger performance toward the weaker performance based on the variable.  Then,

If past performance < partisanship, projection = performance - 2%

If past performance > partisanship, projection = past performance - 2/3 difference with partisanship.

IV.  Long termers (3 or more)

Take weakest of past 2 performances.

If < p-ship, projection = performance - 2%.

If > p-ship, projection = interpolation from past performance back toward p-ship based on variable.

Projection of party and margin

The projection is a percent (0-100) referring to likely Dem vote.

Based on that, we project a party and range (landslide, comfortable, competitive or no projection)

Scheme for projection of party:

                   Incumbent Party

              Projection D          R   

                                    >= 50 + 3           D               Vulnerable

                                   <= 50 – 3            Vulnerable        R        

                                     between             No proj      No proj  
  This area in yellow shows how the projection number (0-100) and incumbent party combine to make a projection about which party will win the seat.

Scheme for projected winning range:
                                   Projected winning party

                          Projection  No proj   Vulnerable   D/R        
                          >= 50 + 10  No proj   vulnerable   Landslide 

                          >= 50 + 5   No proj   vulnerable   Comfortable

                          >= 50 + 0   No proj   vulnerable   Competitive

For example, a Dem in a 57% projection seat is a D projection.  A Dem in a 45% seat is classified as vulnerable.  There is one I seat (Bernie Sanders, VT), and he is treated like a Democrat.

For example, if a Dem is projected (ie, projected party is D and not No Projection or Vulnerable), the projected range is correct under the following conditions:    
party         Projection                     Correct if 
Dem    Competitive    Dem wins

Dem    Comfortable    Dem % >=55%
Dem    Landslide      Dem % >=60%

For example, a Dem in a 57% projection seat is a D projection.  A Dem in a 45% seat is classified as vulnerable.  There is one I seat (Bernie Sanders, VT), and he is treated like a Democrat.

The same patterns applies for Rep seats.  For example, a Rep. "comfortable projection" is correct if the candidate wins with at least 55% of the vote.