2010 Senate Forecast Simulation Model

The 2010 Senate Forecast Simulation Model determines a forecast probability distribution of GOP gains using sensitivity analysis and Monte Carlo simulation. The model is based on the latest Senate polling. It assumes that the election is held today and will be updated periodically to Election Day. The GOP is poised to make significant gains in the House and Senate.

Projecting a mix of RV and LV polls and allocating 60% of the undecided vote to the challenger:
The GOP expected net gain is 6.3 seats (average based on Monte Carlo simulation of 200 trials).
The Democrats retain control of the Senate (51-47).
The GOP wins the popular vote by 50.4-49.6%

Projecting only LV (mostly Rasmussen) polls:
The GOP expected net gain is 8.3 seats (average based on Monte Carlo simulation of 200 trials).
The Senate is tied at 49-49.
The GOP wins the popular vote by 51.5-48.5%

The following equation has proven to be quite accurate in 2004, 2006 and 2008 (see the examples below).
Democratic recorded share = RV Poll + Undecided voter allocation (UVA) - Fraud Component

Democratic projected recorded share = 48.5 = Projected RV share (49.6) - Fraud
Fraud = 1.1% (reduction in Democratic share).

The reduction results in an increase of two (2) GOP seats (from 47 to 49).
For a 2% reduction, the GOP wins the Senate: 50-48.

Note that the equation does not account for the disenfranchised (mostly Democrats) who never even get to the polls.

The most important factor in any election is voter turnout. High turnout is always good for the Democrats. Pre-election registered voter (RV) polls were superior estimates of the True Vote (not the recorded vote) in 2004, 2006 and 2008. Undecided voter allocation is another key variable in election forecasts. The base case assumption is that the challenger will win a majority of undecided voters. The sensitivity analysis table displays the effects of various allocation scenarios.

In the model, the RV poll projection is considered to be the best estimate of the True Vote. The LV projection is considered to be the best estimate of the recorded vote. Due to uncertainty in the undecided vote and the potential for vote-switching, GOP net gains are calculated over a range of assumptions and displayed in the sensitivity analysis tables.

Election fraud is a key factor that is overlooked in polling models. The Fraud Factor is defined as the percentage deviation from the True Vote. The base-case assumption is that the election will be fraud-free (i.e. vote switch is zero). Based on the historical record of state recorded vote deviations from the unadjusted exit polls this is an unrealistic scenario.

Example 1: 2004
Kerry led the final RV polls by 48-47%. After allocating undecided voters, he was projected to win by 51-48%.
Bush won the recorded vote: 50.7-48.3%. See the 2004 Election Model Projection.

Substituting into the formula:
Kerry recorded share = 48.3% = RV Poll + UVA - Fraud = 48 +3 - Fraud
Fraud = 2.7% (deviation from the RV poll projection).

The National Exit Poll indicated that Kerry won by 51-48%, matching the RV Poll projection.
The aggregate of the unadjusted state exit polls indicated that Kerry won by 52-47%.
Kerry won the True Vote by 53.5-45.5%. See the True Vote Model.

Example 2: 2006 Midterms
In the 2006 Midterms, the Generic Poll Trend Model allocated 60% to the Democrats. The linear regression trend projected a 56.4% Democratic share.
The unadjusted 2006 National Exit Poll was an exact match to the projection: 56.4%.
But the CNN Final Exit poll was forced to match the recorded vote: the Democratic share was reduced to 52.6%.
Unadjusted exit polls have not been released.

Projected recorded share = RV Poll + Undecided voter allocation (UVA) - Fraud Component
Democratic share = 52.6 = 52.0 + 4.4 - Fraud
Fraud = 3.8%

120 Generic Poll Linear Regression Trend
Dem = 46.98 + .0419x
Rep = 38.06 + .0047x

Substituting x = 120 and allocating 60% of the undecided vote (UVA) to the Democrats:

Trend + UVA = Projection
Dem = 52.01 + 4.42 = 56.43%
Rep = 38.62 + 2.95 = 41.57%

Example 3: 2008
Obama led the final RV polls by 53-40% - but they were not listed among the final RCP pre-election polls.
He was projected to win by 56-41% after UVA. He had 57% in the True Vote Model.
Obama led the LV polls by 50-43%. He was projected to win by 53-44%, a close match to the recorded vote (52.9-45.6%).
The unadjusted exit polls have not been released.

Projected recorded vote share = RV Poll + Undecided voter allocation (UVA) - Fraud Component
Obama recorded vote = 53 = 53+ 3 - Fraud
Fraud = 3%

Link to the original URL: http://richardcharnin.com/2010SenateForecastSimulationModel.htm

ENDS

© Scoop Media