Reply #3: 2004 EC: '04 Prelim Exit Poll Vote-shares applied to 2008 EC "Est 'Voted-2004' Mix" + implications: [View All]

[tiptoe]

The Election Calculator (EC) is an election-forecasting model that uses demographics (concepts and assumptions) to generate a feasible dataset for projecting an election's Popular Vote Share. It is an independent alternative to the Election Model (EM) reliance solely on polls for its Vote Projections, i.e. State polls, for generating an "Expected Electoral Vote" using Monte Carlo EV Simulation and National polls, using moving averages of the "latest-five" from a collection of national pollster results -- thereby less subject to the vagaries between single polls (types-LV/RV,sizes,methods,erros,etc) -- for projecting Popular Vote shares.

The EC model and the EM model are independent models, using distinct methods and different data sets, and affording independent projections of an election's Popular Vote. They produce results which can be compared.

The 2004 Election Exit Polls generated controversy, when three preliminary exit polls showed Kerry leading by 3% over Bush throughout election night, as the polls closed and before the vote-counts from each state began emerging, East coast to West. (A "computer glitch" disrupted the third, 12:22am Preliminary exit poll results from appearing in official report format, but the Washington Post managed to get the data into print.) The fourth and Final exit "poll", though -- released the day after the election and with its preliminary-demographics "adjusted", having been forced to match the states-reported vote-count (the standard operating practice of pollsters) -- showed Bush "officially" "winning" by a 3% margin, the same margin that Kerry had led by throughout the election-night's three preliminary-polling measures...i.e. a 6% reversal of fortune, not only for American soldiers sent "half-strength" into an unwinnable war in Iraq, based on lies, but also for 1.2 million Iraqis killed and another million injured (more than 4% of Iraq's population).

Which was trustworthy in 2004: the states' recorded vote counts which established Bush the "official" "winner" and to which the Final exit poll was forced to match...or one or more of the three Preliminary exit polls of Interviewed-Voters' responses, as they exited voting booths, and which had Kerry leading Bush throughout Election night?

The 2008 Election Calculator (EC) projected a %-Mix of "2004-voters-returning-to-vote-in-2008" using a mortality assumption, a 95% turnout assumption and an estimation of "new voters" based on the difference between a projection of 143.7 million votes to be cast today and the estimated total for "2004 returning voters". Then, as a test, the 2008 EC applied its calculated %-Mix-weights to the Kerry-Bush-Other vote-shares that appeared in the 2004 12:22am Preliminary NEP 'Voted-in-2000' category to see what would result as the projected popular vote share for Obama/McCain in 2008.

The independent 2008 Final Election Model (EM) projected Popular Vote shares of
Obama-54.3 McCain-44.7% Other-1.0%

The independent 2008 Election Calculator Model (EC) projected Popular Vote shares of
Obama-54.5 McCain-44.4% Other-1.1%

The two 2008 models independently match (within 0.2%) in their Popular Vote Share projection for the same election: The polls-based 2008 Election Model (EM), therefore, gives independent support to the vote shares used in the demographics-based 2008 Election Calculator (EC) "test".

Those votes shares happen to be the same vote shares measured in the 'voted-in-2000' category of the 2004 12:22am Preliminary Nat'l Exit Poll (13,047 random-sample, 1% MoE)

Those same Preliminary NEP vote shares established Kerry winning the True Vote in 2004 by a 66.9–57.1 landslide. (see '12:22am Vote share' in 2004 Calculated True Vote' here )

The 2008 Election Calculator Model confirms the 2004 and 2008 Election Model (and vice-versa)

In May 2008, the 2008 Election Calculator projected that Obama would win the True Vote by 71–59m (54.1–44.7%).

Checking the 2004 Election Calculator (EC) True Vote and the 2008 Election Model (EM) Projections

 

On Nov.3, 2008 the following test was performed:

The 12:22am 2004 NEP vote shares were input to the 2008 EC.
In the 2008 EM, 75% UVA and 3rd party 1% share were input to match 2004 EC assumptions.
The resulting 2008 EC calculated True Vote projection closely matched the EM Polls-projection (to within 0.2%).
Therefore, the EC 2004 vote shares and weighting mix are also confirmed and therefore must be fairly accurate.
The 2008 EC could only be accurate (and match the EM), if the input estimate of returning 2004 Bush and Kerry voters was also accurate.
The model estimates 60m returning Kerry voters and 51.6m returning Bush voters.
Given a 75% UVA and 1% to Other, the EC projects Obama will win by 78.3–63.8 million votes, assuming a fraud-free election.
Note that the base case EM is 60% UVA and 2% Other

2008 Election Calculator (EC)

2004-Voters  Return
in 2008  — Estimate

 

True 'Voted 2004'
Mix  Calculated

12:22am NEP Shares ('13047')

In 2004

Kerry
Bush
Other

125.7

Turnout '08
DNV
95%
95%
95%

113.7

 

Votes
29.9
60.6
51.6
1.6

143.7

Mix
20.8%
42.2%
35.9%
1.1%

100.0%

Obama
57%
91%
10%
64%

54.5%

McCain
41%
8%
90%
17%

44.4%

Other
2%
1%
0%
19%

1.1%

75% UVA

143.7

78.3

63.8

1.5

 

 

2008 Election Model (75% UVA) (EM)

54.3%

44.7%

1.0%

 

143.7

78.0

64.3

1.4

 

2008 Election Model (60% UVA) (EM)

53.1%

44.9%

2.0%

 

143.7

75.9

64.2

2.9