Swing vs. Red-Shift: Killing the “Zero slope, no election fraud” Canard
Richard Charnin
Nov. 21, 2011
http://richardcharnin.com/SwingVsRedshift1992to2004.htm
1/20/12 Update: The 1988-2008 state unadjusted exit polls have recently been made available on the Roper Site. They indicate that
-Obama had 58.0% in the unadjusted state exit polls and 61% in the unadjusted National Exit Poll (exactly matching the True Vote Model).
-Kerry won by 51-47.6% in the unadjusted state exit polls and had 51.7% in the unadjusted National Exit Poll (2% lower than the True Vote Model).
-Gore won by nearly 6 million votes (50-44% (a near match to the TVM). The exit pollsters indicated a 2.0% WPD, but it was closer to 6% based on the unadjusted numbers.
-Dukakis in 1988 led the 24 battleground states unadjusted exit polls by 51-47% and may very well have won the election. He lost by a 53-45% recorded margin (7 million votes).
After the 2004 election, exit pollsters and naysayers claimed that the near-zero correlation between Swing (change in vote share from 2000 to 2004) and Red shift (the 2004 exit poll discrepancy) “kills the fraud argument”. They provided a swing vs. red-shift scatterchart of 1250 precincts and pointed to the flat regression-line as evidence:
http://img303.imageshack.us/img303/3831/swingshift2zb.jpg
But there was a problem with the data. It reflected the recorded vote – not the True Vote (based on votes cast). The pollsters used bogus recorded vote data to prove that there was no fraud – a circular argument if there ever was one. The implicit assumption was that the recorded vote reflected the True Vote, yet there were nearly six million uncounted votes in 2000 and four million in 2004. That fact alone is proof that there is always a discrepancy between the True and recorded vote.
On the other hand, when unadjusted state exit polls are used as a proxy for the True Vote, there is a strong positive correlation. To use the recorded vote as the basis for claiming the 2004 election was fraud-free is an example of the disinformation campaign to cast doubt on the state and national exit polls which indicated that Kerry won by 52-47%.
In fact, Edison-Mitofsky themselves refuted their “no-slope, no fraud” argument in the 2004 Exit Poll Evaluation Report. A state exit poll WPE correlation analysis showed a relatively high 0.48 correlation for 2000-2004. Conversely, the correlation was a much lower 0.05 for 1996-2000. It was 0.15 for 1992-1996 and 0.29 for 1988-1992. If anything, the analysis shows that the 1992 and 2004 elections (Bush Sr. and Jr. were incumbents) were fraudulent, while 1996 and 2000 (Clinton was the incumbent) were fair.
Year 2000 1996 1992 1988
2004 0.48 0.19 0.35 0.30
2000 ……. 0.05 0.12 0.23
1996 ……. ……. 0.15 0.26
1992 ……. ……. ……. 0.29
The 1992-2004 Swing Redshift graph displays True and Recorded Vote Swing and Red shift (based on 238 state exit polls) for the 1992-2004 elections.
http://richardcharnin.com/TrueVoteElectionCalculator_19922004_image001
The 0.39 True Vote Swing/Red shift correlation is significantly higher than the near-zero 0.05 Recorded correlation. The flat regression is partly due to 40 million uncounted votes in the 1988-2004 elections. But there were also millions of votes switched in cyberspace.
Recorded vote share = True Vote + Fraud
Update Note:
The Swing vs. Red-shift correlation ratios have been updated to conform to the unadjusted state exit polls found on the Roper website.
The average recorded vote swing correlation was .01
The average unadjusted exit poll correlation was 0.46
In the original analysis, the exit poll shares were based on average Within Precinct Discrepancy. Both sets of correlations are given below.
Update: Swing vs. Red shift correlation
(Recorded swing vs. unadjusted exit poll swing)
Method Average 1988 1992 1996 2000 2004 2008
Recorded -0.01 na…..0.30 -0.14 0.03 0.15 -0.38
Exit Poll…. 0.46 na…..0.65 0.10 0.57 0.62 0.38
The latest dataset is based on ACTUAL unadjusted state exit poll data. Swing is the CHANGE in the 2-party unadjusted state exit poll share from the PREVIOUS election 2-party unadjusted share. Red-shift is the DIFFERENCE between the 2-party unadjusted STATE EXIT POLL share and the RECORDED VOTE share in the CURRENT election.
The exit pollster’s initial explanation for the discrepancies was that non-response bias skewed the exit polls – the so-called reluctant Bush responder (rBr). Then came Swing vs. Red shift. Finally, False Recall was promoted to explain the impossible number of returning Bush voters implied by the Final National Exit Poll. In each argument, the recorded vote was used as the baseline, rather than total votes cast. Uncounted votes and an estimate of the True Vote were ignored.
To use a fraudulent recorded vote as the basis for calculating swing and then claim that the near-zero correlation “kills the fraud argument” is a logical fallacy. Elections can be fraudulent or fraud-free regardless of the correlation. In 2000, the exit poll discrepancy (2.0 WPD) was much lower than in 2004 (7.4 WPD), yet the correlation ratios were significantly higher in 2000. The scatter graphs below kill the naysayer 2004 Swing/ Red shift “no slope, no fraud” canard.
The fallacy of the argument is confirmed in each of the 1992, 1996 and 2000 elections. The True Vote Swing/ Red shift correlation is much stronger than the corresponding Recorded Vote correlation (the regression line slope is much steeper). The average absolute True Vote correlation was .33 higher than the Recorded Vote.
In the 1988-2008 elections, Democratic presidential candidates did nearly 8% better in unadjusted exit polls (52-42%) than in the recorded vote (48-46%). The discrepancies were due to a combination of uncounted votes and electronic vote switching. The uncounted vote rate trend has declined, but electronic vote switching has more than taken up the slack.
In every election, the swing/red-shift correlations (based on exit poll swing) were significantly higher than the corresponding correlations based on recorded vote swing.
The recorded state vote swing/red shift regression lines were flat in both 2004 (high WPD) and in 1996 (low WPD). On the contrary, in 2000 low WPD) the regression line was steep – refuting the basic premise of the exit pollster argument.
The following graphs display the State Recorded and True Vote share Swing vs. the unadjusted Exit Poll Red shift (i.e. WPD).
In 1992 the WPD was a high 5.4. The correlations: 0.21 Recorded Vote and 0.40 True Vote. Unadjusted exit poll correlation: -0.20. There were nearly 10 million uncounted votes.
http://richardcharnin.com/SwingRedShift_1992_image001.gif
In 1996 the WPD was a low 1.9. The Recorded Vote correlation was nearly zero (.02). The True Vote correlation was 0.43. Unadjusted exit poll correlation: 0.10. There were nearly 10 million uncounted votes.
http://richardcharnin.com/SwingRedShift_1996_image001.gif
In 2000 the exit poll discrepancies (2.0 WPD) were much lower than in 2004. But the 0.38 Recorded vote correlation was higher than the 0.11 in 2004. The True Vote correlation was 0.66. Unadjusted exit poll correlation: 0.57. There were nearly 6 million uncounted votes.
http://richardcharnin.com/SwingRedShift_2000_image001.gif
In 2004, the WPD was a high 7.4. Recorded Vote correlation: 0.11. True Vote correlation was 0.56. Unadjusted exit poll correlation: 0.62. There were close to 4 million uncounted votes.
http://richardcharnin.com/SwingRedShift_2004_image001.gif
In 2008, the WPD was at its highest: 10.3. The regression lines diverged, as indicated by the correlation ratios: -0.38 for Recorded Vote vs. 0.42 for the True Vote. Unadjusted exit poll correlation: 0.60.
http://richardcharnin.com/SwingRedShift_2008_image001.gif
In 2004, the average Battleground state Recorded vote correlation was 0.45; it was near zero in Democratic and Republican states. But the exit poll discrepancies (WPD) in the Democratic states were higher than the Battleground states – another refutation of the premise.
Battleground
http://richardcharnin.com/SwingRedShift_BG_image001.gif
Democrat
http://richardcharnin.com/SwingRedShift_Dem_image001.gif
Republican
http://richardcharnin.com/SwingRedShift_Rep_image001.gif
The Census indicates there were nearly six million uncounted votes in 2000 and 4 million in 2004. Obviously, the recorded vote did not equal the True Vote in either election. It never does. The unadjusted 2004 aggregate state exit poll determined that Kerry was a 52-47% winner, but Bush won the recorded vote by 50.7-48.3%. The aggregate 7.4% WPD is the difference in margin. In 2000, the WPD indicated that Gore’s exit poll margin was 2% higher than his recorded 0.5% margin.
Unfortunately, the National Election Pool consortium has never released unadjusted precinct exit poll data. The transparent claim is the need for exit poll respondent confidentiality. Of course, they could release the data without providing personal information if they wanted to. In their 2004 report, the pollsters did provide average state exit poll WPD data for all elections since 1988. That report provided more than enough historical information to hoist the NEP, the pollsters and the naysayers on their own petard.
