2010 Senate Forecast Simulation Model

Sept. 26, 2010

Richard Charnin (TruthIsAll)

Registered and Likely Voter Projections

The House and Senate forecast models provide Registered Voter (RV) and Likely Voter (LV) projections. The assumption is that the election is held today. Pre-election polls all interview registered voters; likely voters are a sub-sample. They are not separate polls. Democrats always do better in the full RV sample than in the LV sub-sample (see the LVCM model below).

LV polls exclude millions of registered voters who actually vote - and most of them are Democrats. In addition, millions of votes are cast but never counted in every election - and most of them are Democratic as well. The good news is that proliferation of electronic voting has reduced the uncounted vote rate. The bad news is that votes can be switched, stuffed or dropped at the voting machine and/or the central tabulator where they are counted.

Since 2000, LV poll projections have closely matched recorded vote shares and final exit polls (which are forced to match the recorded vote). The RV poll projections closely matched the unadjusted and preliminary exit polls.

Midterm Forecast

(Probability of GOP majority)

Senate Seats

(200 simulated election trials)

LV polls: Dem 50.5 - Repub 47.5 (5%)

RV polls: Dem 52.7 - Repub 45.3 (0%)

The media/pollster drumbeat of a “horse race” is largely based on LV polls. The narrative conditions the public to expect a recorded vote which in fact understates the True Democratic share. The pollsters discount the RV sample for a fraud component, fully expecting that the LV projections will be a close match to the recorded vote - but they never mention the F-word. They know that votes are miscounted in every election. And so their final LV-based polling forecasts are usually quite accurate. Pollsters are paid to predict the recorded vote - not the True Vote.

The 2010 midterms are different from the last four elections in that a low Democratic voter turnout is expected. Election fraud will very likely cost the Democrats a few seats in the House and Senate. And the number will be close to the difference between the RV and LV samples. But there may not be RV samples for us to calculate the difference on Election Day. And once again, pollsters will be complemented on how closely their final LV predictions matched the (recorded) vote.

For the Senate races polling websites generally display only LV polls. CNN/Time provides both RV and LV samples, but only the LVs are listed at realclearpolitics.com. The Senate RV forecast model is therefore a mix of RV and LV polls. The Democrats lead the RV 12-poll average by 47.8-41.1%. The GOP leads the LV 25-poll average by a much bigger margin: 51.6-38.6%. Without a full corresponding RV poll for every LV sample, a comparable analysis is difficult.

In the House, Generic polls have had a more equitable mix of RV and LV samples. But expect a shift to virtually all LV samples as Election Day approaches. The GOP leads the average of 55 LV polls by 45.4-38.5%. For the 83 RV polls the margin is much lower: 45.2-43.7%. The majority of polls are Rasmussen LVs and Gallup RVs.

The Fraud Component

In 2004, 2006 and 2008, projections based on final pre-election LV polls underestimated voter turnout and yet closely matched impossible final exit polls and fraudulent recorded vote counts. Projections based on final pre-election RV polls (adjusted for undecided voters) were a close match to the unadjusted exit polls and the True Vote.

Pre-election Model:

Recorded vote share = LV poll projection = RV poll projection + Fraud component

Post-election Model:

Recorded vote share = Final Exit Poll = Unadjusted exit poll + Fraud component

Applying the formula to the latest Senate and House Generic Polls:

Projected GOP Senate Vote Share:

Share = 52.4 = 50.5 + Fraud component

Fraud component = 1.9% (3.8% margin).

The Likely Voter Cutoff Model (LVCM)

In 2004, there were 22 million voters who did not vote in 2000. Nearly 60% of newly registered voters were Democrats for Kerry. In the 2006 midterms, a Democratic tsunami gave them control of both houses. In 2008, there were approximately 15 million new voters of whom 70% voted for Obama. All pre-election polls interview registered voters. Likely Voter (LV) polls are a subset of the full Registered Voter (RV) sample. LV polls exclude most "new" registered voters – first-timers and others who did not vote in the prior election.

Most pollsters use the Likely Voter Cutoff Model (LVCM), a series of questions regarding past voting history, residential transience, intent to vote, etc. Since students, transients, low-income voters, immigrant new voters, etc. are much more likely to give "No" answers than established, wealthier, non-transient voters, Republicans are more likely to exceed the cutoff than Democrats. A respondent who indicates “yes” to four out of seven questions might be down-weighted to 50% compared to one who answers “yes” to all seven.

The LVCM assigns a weight of zero to all respondents falling below the cutoff, eliminating them from the sample. But these potential voters have more than a zero probability of voting.  The number of "Yes" answers required to qualify as a likely voter is set based on how the pollster wants the sample to turn out. The more Republicans the pollster wants in the sample, the more "Yes" answers are required. This serves to eliminate many Democrats and skews the sample to the GOP.

Undecided Voters, Turnout and Final Exit Polls

In 1988, 11 million votes were uncounted; in 2000, 6 million; in 2004, 4 million; in 2006, 3 million.

In 2004, 2006 and 2008, projections based on final pre-election LV polls closely matched fraudulent recorded vote shares. Projections based on the final pre-election RV polls closely matched the unadjusted exit polls. Undecided voters typically break heavily for the challenger. In each of the last three elections, the Democrats were the challengers, but many pollsters did not allocate accordingly. Democratic voter turnout was underestimated by the pre-election LV polls (see 2004 Final Pre-election polls).

Final exit polls are always forced to match the recorded vote count (i.e. the final pre-election LV polls). The underlying assumption is that the recorded vote is correct (i.e. zero fraud). In 2004 and 2008, the Final National Exit Polls required an impossible turnout of returning Bush voters (110% and 103%, respectivrely). In the 2004 Final (13660 respondents), the Bush vote shares were increased dramatically over the 12:22am Preliminary NEP (13047 respondents). The NEP media consortium of news outlets FOX, CNN, AP, ABC, CBS and NBC has suppressed the release of 2008 unadjusted state exit polls and unforced preliminary national exit polls. 

Once again, as in every election cycle, the media avoids the real issues. Martha Coakley won the hand-counts in Massachusetts for Ted Kennedy’s seat but lost to Scott Brown; Vic Rawl won the absentee vote but lost to unknown Alvin Greene in the South Carolina Democratic Senate primary; Mike Castle won the absentee ballots but lost to Christine O'Donnell in the Delaware GOP Senate primary. But there has not been a peep about any of this in the mainstream media. Apparently, we must just accept the conventional wisdom that even though the votes have vanished in cyberspace and can never be verified, they were not tampered with. The media lockdown is not limited to past stolen elections. The MSM prepares us for election fraud by listing final pre-election LV polls and ignoring RV polls.

Senate Forecast Simulation
http://www.realclearpolitics.com/epolls/latest_polls/senate/
26-Sep		Dem	GOP	Ind	Fraud	0.0%	Vote share deviation
Current seats		57	41	2	MoE	4.0%	Poll margin of error
					UVA	50.0%	Undecided voter allocation to GOP
Polls	Type	Avg Vote Share		Undec
12	RV	47.8	41.1	11.1
25	LV	38.6	51.6	9.8
37	RV/LV	41.6	48.2	10.2
Simulation Forecast (seats)					Projection (table)
37	RV/LV	52.7	45.3	2	RV/LV	Dem	GOP
Net Gain		-	4.3	-	Seats	54	44
Win Probability		100.0%	0.0%	-	Flip to	1	4
Seats	LV	50.5	47.5	2	Lean	4	3
Net Gain		-	6.5	-	Safe	7	16
Win Probability		95.0%	5.0%	-	Tossup	7	0
* tossup			Wtd Share %				Projected %			GOP
Poll	Seat	Type	Dem	GOP	Unsure	Margin	Dem	GOP	Margin	Win Prob		Within
26-Sep	Held	LV	42.5	47.2	10.3	4.7	47.6	52.4	4.7	5.0%	Flip	MoE
Average	By	RV/LV	44.0	45.1	10.8	1.1	49.5	50.5	1.1	0.0%	5	14
AK	R		42	48	10	6	47.0	53.0	6.0	93%		AK
AL	R		30	59	11	29	35.5	64.5	29.0	100%
AR	D		34	54	12	20	40.0	60.0	20.0	100%	GOP
AZ	R		31	53	16	22	39.0	61.0	22.0	100%
CA	D		51	43	6	(8)	54.0	46.0	(8.0)	2%		CA
CO	D*	RV	47	44	9	(3)	51.5	48.5	(3.0)	23%		CO
CT	D		52	45	3	(7)	53.5	46.5	(7.0)	4%		CT
DE	D	RV	59	34	7	(25)	62.5	37.5	(25.0)	0%
FL	R*	RV	34	36	30	2	49.0	51.0	2.0	69%		FL
GA	R		34	52	14	18	41.0	59.0	18.0	100%
HI	D		68	20	12	(48)	74.0	26.0	(48.0)	0%
IA	R		35	55	10	20	40.0	60.0	20.0	100%
ID	R		27	64	9	37	31.5	68.5	37.0	100%
IL	D*		41	44	15	3	48.5	51.5	3.0	77%	GOP	IL
IN	D		29	50	21	21	39.5	60.5	21.0	100%	GOP
KS	R		23	67	10	44	28.0	72.0	44.0	100%
KY	R*	RV	46	46	8	0	50.0	50.0	0.0	50%	Dem	KY
LA	R		33	54	13	21	39.5	60.5	21.0	100%
MD	D		55	39	6	(16)	58.0	42.0	(16.0)	0%
MO	R	RV	41	54	5	13	43.5	56.5	13.0	100%
NC	R		36	56	8	20	40.0	60.0	20.0	100%
ND	D		25	69	6	44	28.0	72.0	44.0	100%	GOP
NH	R		44	49	7	5	47.5	52.5	5.0	89%		NH
NV	D	RV	42	34	24	(8)	54.0	46.0	(8.0)	2%		NV
NY	D	RV	57	31	12	(26)	63.0	37.0	(26.0)	0%
NY	D	RV	63	30	7	(33)	66.5	33.5	(33.0)	0%
OH	R	RV	42	49	9	7	46.5	53.5	7.0	96%		OH
OK	R		24	67	9	43	28.5	71.5	43.0	100%
OR	D		54	37	9	(17)	58.5	41.5	(17.0)	0%
PA	D*	RV	45	45	10	0	50.0	50.0	0.0	50%		PA
SC	R		30	70	0	40	30.0	70.0	40.0	100%
SD	R		30	70	0	40	30.0	70.0	40.0	100%
UT	R		25	52	23	27	36.5	63.5	27.0	100%
WA	D	RV	50	44	6	(6)	53.0	47.0	(6.0)	7%		WA
VT	D		64	29	7	(35)	67.5	32.5	(35.0)	0%
WI	D*		48	46	6	(2)	51.0	49.0	(2.0)	31%		WI
WV	D*		47	45	8	(2)	51.0	49.0	(2.0)	31%		WV
Probability Distribution of GOP Net Gains
LV										Majority
Gain	0	1	2	3	4	5	6	7	8	9	10
Seats	41	42	43	44	45	46	47	48	49	50	51
Probability
Exact	0.0%	0.0%	0.0%	1.0%	5.0%	17.5%	25.0%	37.0%	9.5%	5.0%	0.0%
At least	100.0%	100.0%	100.0%	100.0%	99.0%	94.0%	76.5%	51.5%	14.5%	5.0%	0.0%
Projection Trend
	LV				RV/LV
			Simulation				Simulation
	Vote Share		GOP Gain		Vote Share		GOP Gain
	Dem	GOP	Seats		Dem	GOP	Seats
26-Aug	49.0	51.0	6.2		50.5	49.5	4.4
1-Sep	48.2	51.8	8.0		49.5	50.5	6.0
10-Sep	47.9	52.1	7.3		49.6	50.4	5.3
15-Sep	47.8	52.2	6.7		49.3	50.7	4.3
26-Sep	47.6	52.4	6.5		49.5	50.5	4.3
GOP Forecast Sensitivity to Undecided Voter Allocation and Poll Type
GOP	Vote Share %		Seats (table)			Seats (simulation)		Net Gain
UVA	LV	RV/LV	LV	RV/LV		LV	RV/LV	LV	RV/LV
40%	51.3	49.5	47	42		46.3	43.5	5.3	2.5
45%	51.8	50.0	48	43		47.0	44.3	6.0	3.3
50%	52.4	50.5	48	44		47.5	45.3	6.5	4.3
55%	52.9	51.1	48	46		48.0	46.1	7.0	5.1
60%	53.4	51.6	48	46		48.6	46.7	7.6	5.7
GOP Forecast Sensitivity to Undecided Voter Allocation and Vote Switch
		RV/LV
Vote switch		Undecided Voter Allocation to GOP
to GOP	40%	45%	50%	55%	60%
		GOP Net Gain
0%	1	2	3	5	5
1%	3	5	5	7	8
2%	6	8	8	8	9
3%	8	8	8	10	10
4%	10	10	10	12	12
		Total Seats
0%	42	43	44	46	46
1%	44	46	46	48	49
2%	47	49	49	49	50
3%	49	49	49	51	51
4%	51	51	51	53	53