His disinformation contract should have been terminated by now.

Mark Lindeman is still defending his "false recall" argument while at the same time claiming that I was the one who suggested that he basedit on Bush being 1) a war president and 2) the "winner" of the 2000 election. That is pure DISINFORMATION. Those were the very same arguments that Lindeman suggested in his attempt to explain what he called the "slow moving fog" which caused exit poll respondents to misstate their past votes. Mark was the first to mention the retrospective bandwagon effect in his paper "Too Many Bush Voters: False Recall in the 2004 Election". The fictitious "bandwagon" implies that Bush incumbency and promise to "keep us safe" after 9/11 were factors in "false recall"- despite the fact that the Bush approval rating stood at 48% and the Iraq War was and still is a disaster.

In fact, the NES surveys show that in EVERY election since 1988, the retrospective Democratic margin was HIGHER than the recorded share, CONFIRMING the True Vote Model analysis. The average NES Democratic margin was 14.7%; the average True Vote margin was 15.7%! The average recorded Democratic margin was 6.2%.

Comparing the True Vote average margin to the NES average survey margin for the 11 elections from 1968-2008 debunks "false recall". Why "false recall" was even suggested is a mystery. Apparently, Lindeman never considered that the NES study would CONFIRM the True Vote Model because he never did the analysis.

Here are a few reasons why the "false recall" hypothesis to explain the impossible Final 2004 National Exit Poll 43/37% returning Bush/Gore voter weightings is just false.

1. It assumes that Alzheimer's was strictly a Gore malady; Bush was unaffected.

2. Comparing the difference between the recorded vote and the retrospective past vote is a false dichotomy. The recorded vote, by definition, cannot represent the True Vote since it does not include uncounted votes (a combination of spoiled, absentee, late and provisional ballots) which are heavily (70-80%) Democratic. Only by measuring the past vote against the True Vote (i.e. Total Votes Cast) is the analysis remotely valid.

In fact, the NES surveys show that in EVERY election since 1988, the retrospective Democratic margin was HIGHER than the recorded share, CONFIRMING the True Vote Model analysis. The average NES Democratic margin was 14.7%; the average True Vote margin was 15.7%! The average recorded Democratic margin was 6.2%.

3. False recall is based on the premise that the returning Bush/Gore voter split (43/37%) in the Final National Exit Poll was an actual sampled result and that 3 of 40 (7.5%) Gore voters misreported their past vote. But it is standard operating procedure for the exit polsters to FORCE the Final to match the recorded vote by adjusting selected weights and/or vote shares for ALL demographic categories.

4. Even if we assume that the Final NEP (13660 respondents) returning Bush/Gore voter split was a sampled result, how does one explain the fact at 12:22am (13047 respondents) the split was 41/39%? Surely the 4% change in weights of so few respondents is implausible. Only 3182 respondents were even asked the question about their past vote.

5. The other NEP 10478 respondents were only asked who they voted for MINUTES BEFORE. And 51% said Kerry Surely five minutes was too short a period for "mere forgetfulness", "slow-moving fog' or "bandwagon effect" (Bush had a 48% approval rating). The margin of error for a 10,000 sample is less than 1%.

6. In addition to the Final NEP there were 50 state exit polls. The unadjusted aggregate vote share was 52-47% in favor of Kerry. False recall was proposed to explain the Final NEP past vote anomaly, but ignores the state exit polls in which the respondents were asked who they voted for just MINUTES BEFORE. The Margin of Error for the total 70,000 state sample is significantly lower than 1% - even after adding an incremental "cluster effect" to the theoretical MoE.

7. The 2006 midterm and 2008 Final NEP effectively buried "false recall". Of course, we expected that the returning Bush/Kerry returning voter weights would be ADJUSTED TO MATCH the recorded vote; it's SOP. But look at HOW they were adjusted. In 2006 the BUsh/Kerry mix was changed from 46.5/45.5% in the unadjusted exit poll to 49/43% in the Final. Bush had a 35% approval rating in 2006. The 2008 NEP 46/37% returning Bush/Kerry mix was mathematically impossible; it implied that there were 12 million more returning Bush than Kerry voters. Bush "won" in 2004 by a fraudulent 3.0 million vote margin.

8. To say that a "slow-drifting fog" was unique to Gore voters only is laughable and should be dismissed out-of-hand. Lindeman iss therefore left only with "a latent Bush bandwagon effect" and an "incumbency advantage" as possible explanations. But how come the 12:22am NEP indicated that 10% of returning Bush voters defected to Kerry and only 8% of Gore voters defected to Bush - a counter-intuitive result. Regardless of the rspondents past vote, Kerry enjoyed a net defection in his favor. It strains credulity for Gore voters to vote for Kerry and lie about their past vote. It also strains credulity that these angry Gore voters would FORGET that the election was stolen from THEM when the Supreme Court nullified the Florida Court and halted the recount therby handing the election to Bush.

9. In fact, it was IRRELEVANT whether or nor Gore (or Bush) voters misreported their past vote. Regardless of HOW Gore voters SAID they voted, the only thing that matters is a) how they ACTUALLY voted in 2000 and b) how they ACTUALLY voted in 2004. It is indisputable thatt Bush had 50.5 million recorded votes and Gore had 51.0 million. We will ignore the FACT that were approximately 3.5 million net uncounted votes and that 70-80% of them were for Gore. The recorded 2000 vote and the 1222am NEP indicated that Gore AND Kerry were the winners. PERIOD. The Final NEP weightings do not reflect the demographic realities of Bush and Gore voter mortality and turnout. Even if RETURNING Bush LIVING VOTER TRURNOUT was 98% and Gore LIVING VOTER TURNOUT was 90%, Kerry still wins the calculated True Vote by 7 million.

10. Finally, consider the "The Game", the Aug, 2005 Democratic Underground online debate between spreadsheet-wielding Internet bloggers and the two most prominent exit poll naysayers Lindeman and Elizabeth Liddle. Prior to the "Game", Lindeman was challenged time and again to provide a plausible scenario to to show how Bush "won" the election. Lindeman stipulated that we are all mortal and reduced the number of returning Gore and Bush voters accordingly by the same mortality rate. He also assumed equal Bush and Gore voter turnout rates.

So far, so good. Lindeman effectively abandoned "false recall" to provide a feasible, plausible returning voter mix. Unfortunately, he could NOT match the 2004 recorded vote using the Final NEP Bush vote shares which had ALREADY BEEN INFLATED TO MATCH THE VOTE. His only recourse was to INFLATE the shares even further to IMPLAUSIBLE LEVELS, FAR BEYOND THE MARGIN OF ERROR. The "Game" effectively proved that he could not provide ONE PLAUSIBLE BUSH WIN SCENARIO.

But Lindeman did not quit; he had to revert back to "false recall" to explain the 2004, 2006 and 2008 Final NEP returning Bush voter anomalies. And we know how that worked out.

http://www.democraticunderground.com/discuss/duboa...

OnTheOtherHand (1000+ posts) Sat Feb-14-09 07:17 AM
Response to Reply #1
8. get used to it

Edited on Sat Feb-14-09 07:54 AM by OnTheOtherHand

I won't say that TIA actually intends not to make sense. But I'll say that he has a particular knack for incomprehensible tables, and that it's common for people to make comments along the lines of "I don't know exactly what this means, but it supports what I already believed."

ETA: Basically, this seems to recapitulate errors that TIA has made repeatedly in interpreting the 2004 results. TIA's analysis here relies heavily on people's reports of their past votes. For instance, in the weighted 2008 exit poll, roughly 46% of respondents said they voted for Bush in 2004, and 37% that they voted for Kerry. (MSNBC tables) But Bush didn't beat Kerry by nearly that much. For TIA, that means that the pollsters must have created lots of extra McCain voters, and thereby, lots of extra Bush '04 voters.

However, as I point out here and in more detail here (PDF), previous winners -- Democrats and Republicans alike -- almost always do better in the retrospective question ('who did you vote for four years ago?') than they did in the official returns. This is true not only in the exit polls, but in the two best-known academic surveys (ANES and GSS).

So, TIA is wrong from point 1. He says that to believe that Obama won by 9.5 million votes, I must also believe that Bush won in 2004 by 12.5 million votes. If I thought that the 46% and 37% figures were accurate, that might be true. But decades of polling experience indicate that those figures are very unlikely to be accurate. So, I don't have to believe that Bush won by 12.5 million votes, and I don't.

As to how many votes Obama won by, I'm open to serious argument. TIA's rehash of old mistakes doesn't pass muster.

He goes on:

OnTheOtherHand (1000+ posts) Sat Feb-14-09 09:43 PM
Response to Reply #12
14. fact check

TIA
To deal with the 43/37 conundrum, exit poll naysayers had to come up with another hypothesis — "false recall": Returning-Gore-voters misspoke (lied, forgot) and told the exit pollsters they voted for Bush in 2000 because they wanted to associate with his win of the 2000 (stolen) election, and he was a war president (who ignored warnings of 911) and had a 48% approval rating to show for his efforts.

OTOH
As far as I know, the assertion that respondents "wanted to associate with (Bush's) win of the 2000 (stolen) election" was invented by TIA himself. Nevertheless, the evidence of routine overstatement of support for a previous winner -- in survey after survey, election after election, exit polls and other polls -- is extensive. I've linked to some summaries in post #8 for your convenience. TIA never has offered a satisfactory response to this evidence, so I don't expect one now.

TIA
To suggest that the 46/37 Bush/Kerry returning-voter mix was due to Kerry voter "false recall" that they voted for Bush (at 22% approval in '08) makes absolutely no sense...

OTOH
TIA, to be charitable, has managed to misunderstand the argument all this time, so I guess it's not surprising that he still has failed to examine or to comprehend any of the historical evidence. It is TIA's assertion alone that false reporting of past votes depends on the popularity of the president in question, although there may be some correlation. If anything, the magnitude of the gap between retrospective reports and the previous results seems to depend more heavily on the prominence of the losing candidate. John Kerry was not especially prominent after the 2004 election, so his (roughly average) retrospective decline in the exit poll isn't surprising.

It isn't a matter of "getting hit by lightening (sic) two days in a row." As I've pointed out to TIA repeatedly, this phenomenon has occurred in every presidential exit poll in the ICPSR archives.
----------------------------------------------------------------------------------

TIA

I will say that OTOH actually intends not to make sense. That he has a particular knack for incomprehensible verbiage. That it's common for readers to comment along the lines of "I don't know exactly what this means, but it is contrary to common sense."

Basically, OTOH just recapitulates arguments that he has made repeatedly in attempting to explain the impossible Final 2004 National Exit Poll return voter weightings: that it was due to respondent "false recall". OTOH relies on public naivete regarding unadjusted "pristine" and adjusted "bogus" final exit polls. He avoids the fact that the exit pollsters themselves admit they always force a match to the recorded vote.

Exit pollsters can only force a match to the recorded vote by adjusting returning voter weights and/or vote shares. The "Voted 2000" category weights in the 2004 Final NEP 43/37 Bush/Gore weights were impossible. If weighting adjustments are not sufficient to force a match (as in 2004) then the vote shares are changed as much as necessary to finish the job of matching to the vote. Or vice-versa. Compare the 1222am NEP vote shares and returning voter weights to the Final NEP.

OTOH also misrepresents the results of the pre-election polls by failing to acknowledge that pre-election Registered Voter (RV) polls are superior to Likely voter (LV) polls - especially when there is a heavy turnout of new voters like in 2004. The unadjusted 2004 exit polls matched the final RV pre-election polls (after 70-80% of undecided voters were allocated to Kerry).

The exit pollsters/media have not released the 2008 exit poll report which would reveal the average within precinct discrepancy (WPD) for each state from which one can calculate the aggregate national share. WPD is the difference in margin between the exit poll and recorded vote.

The 2008 True Vote Model confirmed the pre-election RV polls. That alone is powerful evidence that the Obama landslide was denied - big time. OTOH concedes that the Final 2008 National Exit Poll result that 46% of respondents voted for Bush in 2004 and 37% for Kerry is unlikely because Bush didn't beat Kerry by nearly that much. In fact, Kerry beat Bush, which makes the 46/37 split even more outrageous.

According to OTOH, the 46/37 anomaly is just another example of returning Kerry voters misstating their past vote. OTOH cannot accept that in order to match the recorded vote the pollsters created lots of extra McCain voters, and thereby, lots of extra Bush '04 voters.

OTOH is once again exposed by his own misinformation. He concludes in his paper "False Vote Recall and the 2004 Exit Poll" (see comments 8-12 later in this thread for the full text) that respondents "forget" who they voted for in the previous election due to " a slow-drifting fog" and/or a retrospective "incumbency advantage".
------------------------------------------------------------------------------

His conclusion:
"False vote recall complicates our analysis of partisan dynamics, and challenges some unconscious assumptions. Political observers rarely profess surprise that some respondents wrongly claim to have voted, but some find it strange that millions of voters might misreport – indeed, might forget – whom they voted for four years ago.

I have not presented (or uncovered) systematic evidence about the mechanism behind false vote recall, but mere forgetfulness is not a bad account for respondents who (e.g.) report in 2000 that they voted for Gore, then report four years later that they had voted for Bush in 2000 but for Kerry in 2004. No spiral of silence this: more like a slow-drifting fog.

I am reminded of Larry Bartels’ (1996) conclusion that presidential incumbents derive approximately a five-point advantage from information effects” (or, one might say, non-information effects) in the electorate. False vote recall favoring the previous winner is one distinctive manifestation of this incumbency advantage, although its practical importance is difficult to gauge – especially given the confounding influence of differential turnout".

As far as his statement that "previous winners -- Democrats and Republicans alike -- almost always do better in the retrospective question ('who did you vote for four years ago?') than they did in the official returns. This is true not only in the exit polls, but in the two best-known academic surveys (ANES and GSS)"
-------------------------------------------------------------------------------

So much for his veracity in claiming that these arguments were mine and not his. But what would you expect from such a premier disinformationist - the Truth?

He says: "TIA, to be charitable, has managed to misunderstand the argument all this time, so I guess it's not surprising that he still has failed to examine or to comprehend any of the historical evidence."

Mr. Disinformation accuses me of misunderstanding? That's rich. It's a typical Rovian ploy to accuse your adversary of what you yourself are guilty of. No, Mr. Hand, you're the one who is not telling the truth - not the exit poll respondents.

In fact, the NES surveys show that in EVERY election since 1988, the retrospective Democratic margin was HIGHER than the recorded share, CONFIRMING the True Vote Model analysis. The average NES Democratic margin was 14.7%; the average True Vote margin was 15.7%! The average recorded Democratic margin was 6.2%.

One More Time, Mr. Hand:
YOU FAIL TO UNDERSTAND (OR ADMIT) THAT THE FINAL NEP RETURNING VOTER MIX DOES NOT REFLECT WHAT RESPONDENTS SAID ABOUT THEIR PAST VOTE. THE WEIGHTINGS ARE ARBITRARILY ADJUSTED IN THE FINAL NEP TO FORCE IT TO MATCH THE RECORDED VOTE - EVEN IF IMPOSSIBLE WEIGHTINGS ARE REQUIRED. COME HELL OR HIGH WATER. IT'S SOP. IT'S THE ONLY WAY THAT THE MSM CAN COVER-UP THE FRAUD - BY REPLICATING IT.

YOUR ARGUMENT IS EXPOSED BY YOUR "BELIEF" THAT BUSH WON BY 3 MILLION VOTES IN 2004. NO ONE CARES WHAT YOU "BELIEVE", ONLY WHAT YOU CAN DEMONSTRATE. YOUR EFFORTS HAVE JUST PRODUCED FOUR YEARS OF UNRELENTING FOG WITH JUST ONE INTENT - TO MISINFORM THE PUBLIC ABOUT THE PRE AND POST-ELECTION POLLS.

THE "ONTHEOTHERHAND" MONIKER GIVES YOU AWAY, MR. NO. OF COURSE, WITH YOUR CONNECTIONS, WHY SHOULD YOU CARE? ALTHOUGH YOUR ARGUMENTS HAVE BEEN REFUTED TIME AND AGAIN, YOU STILL PERSIST AND YOU KEEP COMING BACK - JUST LIKE RETURNING PHANTOM BUSH VOTERS IN 1988, 1992, 2004, 2006, 2008.

A NATIONAL EXIT POLL ANALYSIS SHOWS THAT MORE THAN 100% OF PREVIOUS NIXON, BUSH 41 AND BUSH 43 VOTERS STILL LIVING WAS NECESSARY IN ORDER TO MATCH THE RECORDED VOTE. THAT'S ANOTHER NAIL IN THE COFFIN OF THE "FALSE RECALL" HYPOTHESIS.

THERE IS ONLY ONE WAY THAT OTOH CAN EXPLAIN THE 2008 NATIONAL EXIT POLL ANOMALY OF 12 MILLION MORE RETURNING BUSH THAN KERRY VOTERS - TO ONCE AGAIN USE THE SPECIOUS ARGUMENT THAT IT WAS DUE TO EXIT POLL RESPONDENTS MISSTATING THEIR PAST VOTE.

OTOH, THAT DOG WON'T HUNT.

WE KNOW THAT THE MIX WAS AN ARBITRARY ADJUSTMENT MADE BY THE EXIT POLLSTERS. BY THERE OWN ADMISSION, IT IS STANDARD OPERATING PROCEDURE TO MATCH THE FINAL EXIT POLL TO THE RECORDED VOTE. TO OTOH, THERE IS NOTHING SUSPICIOUS HERE.

OTOH ESSENTIALLY IGNORES THE FACT MATCHING TO THE RECORDED VOTE IS NOT ONLY STANDARD OPERATING PROCEDURE - IT'S A SCAM. THE FINAL EXIT POLL DOES NOT REPRESENT WHAT THE EXIT POLL RESPONDENTS SAID ABOUT THEIR PAST VOTE. IT ONLY REPRESENTS THE DUPLICITOUS EXTENT TO WHICH THEY HAD TO GO TO IN ORDER TO MATCH THE RECORDED VOTE.

OTOH, YOU HAVE IT EXACTLY BACKWARDS. THE EXIT POLL RESPONDENTS TOLD THE TRUTH; IT WAS THE EXIT POLLSTERS WHO LIED. THEY HAD TO. THEY WERE PAID TO MATCH THE RECORDED VOTE. WHAT MAKES YOU THINK THE POLLSTERS WOULD DEFY THEIR PAYMASTERS?

WOULD YOU DEFY YOURS?

In this post, I address issues raised in the following paper by Mark Lindeman (OnTheOtherHand)

http://inside.bard.edu/~lindeman/too-many.pdf

Too Many Bush Voters? False Vote Recall and the 2004 Exit Poll
by Mark Lindeman

I have been staring at the exit poll results since Tuesday, and one result stands out more than any other: Presidential vote in 2000. If the CNN exit poll is to be believed, Kerry lost because Gore voters were far less likely to vote on Tuesday than were 2000 Bush voters.[… According to the exit poll,] only 37% of the 2004 electorate voted for Gore in the last election, vs. 43% for Bush…. That result strains credibility. Any thoughts?

Obviously these results do not remotely correspond with any analysis or projection of voter turnout performed either before or after the election. Frankly, they defy belief. I would love to see the results of the exit polling on this question before it was "crosstabulated" to conform with the "actual" results. Until a rational explanation is advanced for this anomaly (assuming that is even possible), I feel justified in reacting with great skepticism to the assertion that the "official" results are reliable while the exit poll results are not.

–excerpts from two comments posted at mysterypollster.com, 11/7/04

Since shortly after the 2004 election, scattered observers have cited the apparent excess of Bush 2000 voters in the weighted 2004 National Election Pool (NEP) exit poll tabulations as evidence of election fraud. (Steve Freeman featured this argument in his joint appearance with Warren Mitofsky in October 2005.2) In brief and extreme form, the argument runs as follows: The exit poll indicates that 43% of 2004 voters voted for George W. Bush in 2000, which would equal about 52.4 million Bush 2000 voters. Bush received fewer than 50.5 million votes in 2000 – and, of course, some of those voters have died. Ergo, the only way to account for Bush’s popular vote margin is to assume that more than 100% of Bush 2000 voters turned out in 2004. Since this is impossible, the parsimonious conclusion is that Kerry won the popular vote.

While this argument has some obvious gaps, it does raise the question: what might account for the apparent overrepresentation of Bush 2000 voters in the 2004 exit poll? The short answer is deceptively simple: in all likelihood, many 2000 non-voters and Gore voters wrongly stated that they had voted for Bush. A corollary is that Gore 2000 voters “defected” to Bush in 2004 at a substantially higher rate than Bush 2000 voters defected to Kerry, although the exit poll tabulation indicates that the rates are very similar.

TIA:
There it is: False recall. In all likelihood? There is no question. By the exit pollster's own admission, they always force the Final National Exit Poll (i.e. adjust the weights and vote shares) to match the vote.

OTOH:
The apparently misleading tabulation of 2004 by 2000 vote not only has been cited as proof of fraud, but has influenced at least one published account of Bush’s victory. Pomper (2005, 50) asserts that “small – and offsetting – proportions of Gore and Bush voters switched sides. Bush won in 2004 because there were more consistent, ‘standpat’ Republicans than Democrats. This group appears as the largest segment…, again demonstrating the mobilization of party loyalists.” While I do not doubt that the Bush campaign mobilized party loyalists effectively, I argue that Bush’s improvement over 2000 probably owes more to differential defection than differential mobilization.

TIA:
And the proof of this differential defection is in “false recall”? In other words, he does not accept the 12:22am NEP which states that 10% of Bush voters defected and 8% of Gore voters - even though the margin of error for both is approximately 1%?

OTOH:
In this paper, I first briefly review the “exit poll debate,” for the benefit of interested readers. I then turn to the topic of false vote recall; review evidence about false recall of past presidential votes; and draw on the 2000-04 National Election Study panel to attempt to assess how false recall has affected the defection rates reported in the 2004 NEP tabulation. I conclude that the defection rate of Gore 2000 voters to Bush was probably about twice as large as the defection rate of Bush 2000 voters to Kerry – although any such conclusion hinges on tenuous assumptions. I similarly find that the exit poll crosstabulation exaggerates Kerry’s margin among people who did not vote in 2000. In any case, past vote reports cannot be relied upon to analyze political dynamics across elections.

TIA: Twice as large? Really. In other words, six times the margin of error with zero probability? With Bush at 48% approval? When you say that past vote reports cannot be relied upon, is Lindeman referring to votes RECORDED or votes CAST? I presume you mean RECORDED votes. There is quite a difference.

OTOH:
The exit poll controversy

The unlikely notoriety, within a small circle, of the 2004 NEP exit poll’s 2004-by-2000 vote tabulation owes to the controversy regarding the results. Shortly after 7:30 pm EST, CNN.com posted preliminary tabulations from the national exit poll that implicitly gave Democrat John Kerry a lead of approximately 3 percentage points. CNN also posted tabulations from the Ohio exit poll that indicated that Kerry led by about 4 points in that decisive state.3 After midnight, the tabulations were updated, primarily to match official returns that showed Bush ahead in Ohio and in the national popular vote. Some observers believed that Kerry had in fact won the popular vote, and the exit poll had been “cooked” to obscure the true result. Some subsequently argued that internal contradictions in the exit polls supported their case (see, for instance, the sources cited in footnote 2).

TIA:
He says “primarily to match official returns”. Ah, there’s the rub. So there is no argument there. Actually, Kerry was leading by slightly more than 3.7% at 12:22am in the National Exit Poll; the preliminary unadjusted results were already in process of being matched to the recorded vote.

Furthermore, although we did not know this until the exit pollsters report was issued in Jan. 2005, the unadjusted state exit poll aggregate (over 100,000 sampled) indicated that Kerry won by 52-47%. The "false recall" analysis refers to 3,200 of 13,047 responders who were asked how they voted in 2000. What about the other 10,000 NEP respondents who were asked who they JUST voted for? And what about the 70,000 state exit poll respondents who were asked who they JUST voted for?

The 2004 True Vote Model is based on total votes cast in 2000 and 2004, voter mortality and a plausible turnout of previous Bush and Gore Voters. In other words, it is based on demonstatable demographics. Kerry won the True Vote by approximately 10 million.




12:22am NEP
Requires 105% turnout of Bush 2000 voters

Voted Mix Kerry Bush Other Turnout/Alive
DNV 17% 57% 41% 2% -
Gore 39% 91% 8% 1% 98.4%
Bush 41% 10% 90% 0% 104.6%
Other 3% 64% 17% 19% 97.5%

NEP 100% 51.20% 47.50% 1.30% 3.70%
Votes 122.30 62.62 58.09 1.59 4.52

Recorded 122.30 48.27% 50.73% 1.00% -2.46%
Votes 122.30 59.03 62.04 1.23 -3.01

Final NEP
(forced to match the the recorded vote)
Requires 110% turnout of Bush 2000 voters

Mix Kerry Bush Other Turnout/Alive
DNV 17% 54% 45% 1% -
Gore 37% 90% 10% 0% 93.4%
Bush 43% 9% 91% 0% 109.7%
Other 3% 71% 21% 8% 97.5%

NEP 100% 48.48% 51.11% 0.41% -2.63%
Votes 122.30 59.29 62.51 0.50 -3.22

Recorded 100% 48.27% 50.73% 1.00% -2.46%
Votes 122.30 59.03 62.04 1.23 -3.01


True Vote 95% 98%
Cast Record Alive Voted 2004 True Vote
2000 2000 2000 2004 2004 Turnout Mix Kerry Bush Other Turnout/Alive
DNV - - - - 22.56 15.6% 57% 41% 2% -
Gore 55.79 51.00 53.00 51.94 51.94 42.5% 91% 8% 1% 98.0%
Bush 51.05 50.46 48.50 47.53 47.53 38.9% 10% 90% 0% 98.0%
Other 3.99 3.96 3.79 3.71 3.71 3.0% 64% 17% 19% 98.0%

Total 110.83 105.42 105.28 103.18 125.74 100% 53.39% 45.30% 1.31% 8.09%
Cast Recorded Total 67.13 56.96 1.65 10.17
Gore 50.34% 48.38% Recorded 48.27% 50.73% 1.00% -2.46%
Bush 46.06% 47.86% 122.30 59.03 62.04 1.23 -3.01
Other 3.60% 3.76%




OTOH:
The 2004 exit poll incorporated a total of some 110,000 interviews nationwide. Most respondents were interviewed as they left polling places after voting on Election Day; these interviews took place at over 1400 polling places. Telephone samples of early and absentee voters in twelve states supplemented the polling place interviews. Most exit poll interviews used questionnaires customized for each state. A subset of the nationwide precinct sample was selected for the national exit poll. Participants in the national exit poll received one of four forms of the national questionnaire. Responses from a total of 12,219 national questionnaires were tallied in the final NEP analysis, but the 2000 retrospective vote question appeared on only one form. Thus, 3182 responses are available to the question, “Did you vote in the presidential election in 2000?”5 The resulting weighted tabulation of 2004 vote by reported 2000 vote appears on CNN.com in approximately the form shown in Figure 1 below.

TIA:
All that is true. But here's the rub: the question is irrelevant. The unadjusted state exit poll aggregate (over 70,000 sampled) indicated that Kerry won by 52-47%. Your "false recall" analysis refers to 3,182 responders who were asked the question "who did you vote for in 2000"? What about the other 10,000 NEP respondents who were not asked about their past vote, but simply "who did you JUST vote for"? And what about the 70,000 who were also asked "who did you JUST vote for?

OTOH:
3 In fact, the Best Geo estimate, based on exit poll interviews alone, for Ohio gave Kerry a 6.5 point margin(Edison/Mitofsky [E/M] 2005, 22).

4 The apparent sample size of the dataset is n = 13,719, but each of 500 telephone interviews is entered four times.(The telephone interviews incorporated questions from all four forms.) The tabulation that appears on CNN.com has a nominal n of 13,660, and thus is based on approximately 12,160 respondents. See E/M (2005, 74).

5 The response categories are, in order, “No, I did not vote”; “Yes, for Al Gore”; “Yes, for George W. Bush”; and “Yes, for another candidate.” All four forms of the national questionnaire are available at
http://www.exitpoll. net/election-night/Nat_Final.pdf
(last accessed 1/9/06).

How are these results interpreted as circumstantial evidence of fraud? As I have noted, since about 121 million Americans voted in the 2004 election, 43% of the electorate would equal approximately 52.4 million voters. Yet George W. Bush received fewer than 50.5 million votes. Allowing for mortality, something under 49 million of these voters could possibly have voted in the 2004 election. Why, then, do the weighted exit poll results indicate the existence of too many Bush (2000) voters? To some, this result seems especially incongruous because, prior to weighting, the exit polls are assumed to incorporate interviews from too few Bush (2004) voters.

TIA:
To be precise, there were 122.3 million RECORDED votes in 2004. Bush received 50.5 million. Approximately 2.5 million died, therefore a maximum of 48 million could have voted - assuming 100% turnout. According to the Final NEP, Bush voter turnout was 110% (52.5/48).

OTOH:
These observers conclude that the unweighted exit poll results and/or preliminary tabulations7 are more accurate than the weighted results – which implies that John Kerry won the popular vote. (The unweighted 2004 data indicate that Bush 2000 [Bush2K] voters were 39.5% of the electorate and Gore 2000 [Gore2K] voters 38.4%, among 3182 respondents. These figures are consistent, within sampling error, with the suppositions that Bush2K and Gore2K voters (1) turned out at roughly equal rates, and (2) participated in the exit poll at roughly equal rates. The preliminary tabulations most often cited by fraud theorists actually show a larger gap. The unweighted results and preliminary tabulations both put Kerry ahead.)8

TIA:
Let's review the 12:22am NEP. Note that even here, the 41/39 Bush/Gore mix was impossible, since 41% of 122.3m is 50.1m. And we have just shown that the MAXIMUM Bush voter turnout was 48 million. Therefore, although Kerry is the winner by 4.5 million votes, his margin must have been several million greater.




Voted Mix Kerry Bush Other Turnout/Alive
DNV 17% 57% 41% 2% -
Gore 39% 91% 8% 1% 98.4%
Bush 41% 10% 90% 0% 104.6%
Other 3% 64% 17% 19% 97.5%

NEP 100% 51.20% 47.50% 1.30% 3.70%
Votes 122.30 62.62 58.09 1.59 4.52

Recorded 122.30 48.27% 50.73% 1.00% -2.46%
Votes 122.30 59.03 62.04 1.23 -3.01



OTOH:
On the contrary, I argue that the most parsimonious explanation of the “too many Bush voters” in the 2004 exit poll is that many voters falsely reported having voted for the incumbent. This explanation has not found warm support among advocates of fraud theories. Before systematically presenting evidence for false reporting of past voting, I challenge the plausibility of the fraud account by applying the same reasoning to the 2000 exit poll.

TIA:
By focusing on the 3182 National Exit Poll responders, Lindeman avoids the 70,000 STATE EXIT POLL RESPONDENTS WHO ALSO CLAIMED TO HAVE VOTED 52-47% FOR KERRY.

In any case, we already KNOW what the DEMOGRAPHIC CONSTRAINTS were on the 2004 National Exit Poll turnout weights: 2000 election total VOTES CAST, adjusted for 5% VOTER MORTALITY and 2000 VOTER TURNOUT IN 2004 WHICH HAD TO BE LESS THAN 100%. You are implying that the 43% returning Bush share is a SAMPLED result. Even if it was, it DID NOT REPRESENT THE DEMOGRAPHIC REALITY. Rather, it had to be the "plug" that was required to MATCH THE RECORDED VOTE. In fact, considering the demographic constraints of the 2000 recorded vote (and even total votes cast), it was MATHEMATICALLY IMPOSSIBLE.

OTOH:
6 This table is derived from the CNN.com report of “U.S. President / National / Exit Poll” results at
http://www.cnn.com/ELECTION/2004/pages/results/states/US/P/00/epolls.0.html.

To enhance readability, the table is pasted in HTML and then reformatted to approximate the original appearance.

7 The “unweighted results” to which I refer are the data archived at
ftp://ftp.icpsr.umich.edu/pub/FastTrack/General_Election_Exit_Polls2004/
without weights applied.

Some observers have referred to the preliminary tabulations released on election night as “unweighted,” meaning that they had not been forced to match the official returns. These tabulations do incorporate a variety of weights.

8 Two independent sources indicate that the preliminary national exit poll tabulation as of approximately 7:30 pm EST (which would not incorporate many late results from the West) indicated that the electorate contained 41% Bush2K voters and 38% Gore2K voters, as well as 4% who voted for another candidate and 17% who did not vote.

http://www.scoop.co.nz/stories/pdfs/Mitofsky4zonedata US2004G_3798_PRES04_NONE_H_Data.pdf;

http://web.archive.org/web/20050207070506/http://www.exitpollz.org/CNN_national2.htm

Consider: In the 2000 VNS exit poll, the weighted results indicate that 45.6% of respondents had voted for Bill Clinton in 1996, while 31.3% had voted for Bob Dole, 6.5% for Ross Perot, 2.4% for another candidate, and 12.5% had not voted (1.7% skipped the question). The unweighted results are similar.9 Thus, Clinton’s apparent margin over Dole is 14.3% of all respondents, or 16.7% of those who said they had voted in 1996. By contrast, in the 1996 returns, Clinton received 49.2% of the vote, Dole 40.7%, Perot 8.4%, and other candidates about 1.7%. That is, Clinton won by only 8.5 points. Of course, we cannot assume that the 2000 proportions should match the 1996 proportions. Nonetheless, these results are “impossible” in precisely the same sense as the 2004 results. In 2000, 105.4 million presidential votes were counted, of whom 45.6% would equal about 48.1 million Clinton 1996 voters. Yet Clinton received only 47.4 million votes in 1996, of whom presumably under 46 million survived in 2000. The maximum possible proportion of Clinton 1996 voters in the 2000 electorate is some two points lower than the reported proportion – a discrepancy well beyond expected sampling error.10

How, then, do we explain the “impossibly high” proportion of Clinton 1996 voters in the 2000 exit poll? Is it likely that Al Gore stole millions of votes in 2000, and Clinton voters were upweighted in the exit poll to match the result? Or that millions of votes were stolen from Clinton in 1996? Perhaps anything is possible, but given that the 1996 and 2000 national exit poll estimates came close to the official returns, neither conjecture seems at all likely. Likelier, again, is that many respondents in 2000 wrongly reported having voted for Clinton in 1996.


TIA
That's a reasonable question. But Lindeman as always, cites the recorded vote- not total votes cast. He ignores uncounted votes. The 2000 Final National Exit Poll as always, was forced to match the recorded vote. The True Vote is based on the unmentionable Total Votes CAST.

In 2000, there were 110.4 million votes cast and 105.4 million recorded.
There were 5.4 million net uncounted votes.
In 1996, there were 105.2 million votes cast and 96.3 million recorded.
There were 8.9 million net uncounted votes.

Let's include uncounted votes in order to determine the True Vote in both elections. The best estimate is that 75% of the uncounted votes were for Clinton. Therefore, there were 54.0 million votes cast for Clinton, 41.2m for Dole, 9.9m for Dole.

Deduct 5% for mortality and assume that 96% of voters turned out in 2000. Then there were 51.4m returning Clinton voters, 35.6m Dole and 8.8m Perot/other and 15.9m DNV96. The resulting True returning voter mix was 48.7% Clinton, 33.8% Dole and 8.3% Perot/other, a 14.9% Clinton margin

The returning Clinton True Vote Margin (14.9%) is within 0.6% of VNS (14.3%)!
That may explain it.





Method: Forced to Match
2000 95% 96%
Cast Record Alive Voted Voter
1996 1996 1996 2000 2000 Turnout Mix Gore Bush Other Turnout/Alive
DNV - - - - 15.92 15.1% 42% 51% 7% -
Clinton 49.2% 53.95 47.40 45.03 43.23 48.07 45.6% 82% 15% 3% 107%
Dole 40.7% 41.16 39.20 37.24 35.75 33.00 31.3% 7% 91% 2% 89%
Perot 10.1% 9.90 9.68 9.20 8.83 8.43 8.0% 31% 61% 8% 92%

Total 105.02 96.28 91.47 87.81 105.42 Total 48.41% 47.90% 3.69%
51.03 50.50 3.89

Recorded 48.38% 47.86% 3.76%
105.42 51.00 50.46 3.96


Method: True Vote
2000 95% 96%
Cast Record Alive Voted Voter
1996 1996 1996 2000 2000 Turnout Mix Gore Bush Other Turnout/Alive
DNV - - - - 15.05 9.1% 42% 51% 7% -
Clinton 53.7% 56.34 47.40 53.53 51.39 51.39 48.7% 82% 15% 3% 96%
Dole 37.2% 39.03 39.20 37.08 35.60 35.60 33.8% 7% 91% 2% 96%
Perot 9.2% 9.64 9.68 9.16 8.79 8.79 8.3% 31% 61% 8% 96%

Total 105.02 96.28 99.77 95.78 110.83 Total 48.76% 47.79% 3.76%
54.04 52.97 3.82

Recorded 48.38% 47.86% 3.76%
105.42 51.00 50.46 3.96



Why not also consider the pristine unweighted, unadjusted state exit polls? You claim they are "unweighted" since they have not been forced to MATCH the FINAL official returns. But at 7:30pm, the 2004 exit poll results were ALREADY IN THE PROCESS OF BEING WEIGHTED (BEST GEO). Therefore the 41/38 weights CANNOT REPRESENT THE PRISTINE, UNADJUSTED AND UNWEIGHTED RETURNING VOTER MIX. Furthermore, the 41/38 mix is ALSO mathematically impossible, based on the 2000 recorded vote although 41% is closer to the FEASIBLE 39% MAXIMUM Bush weighting.

OTOH:
Of course, the aftermath of a presidential assassination does not lend itself to robust generalizations. Are survey respondents in general – and voters in particularly – generally predisposed, in subsequent years, to misreport having voted for a victorious presidential candidate? Surprisingly, I have not yet encountered a paper that addresses precisely this question. The extant literature seems to examine two related questions: the extent to which respondents overreport voting, especially in the immediate aftermath of an election; or the extent of misreporting having voted for the winner soon after an election (sometimes called a “bandwagon effect”).

TIA:
Surely SOME respondents misreport – but why would they misreport JUST for Bush? Didn't his 48% approval rating tend to minimize any possibility of a bandwagon effect? Why is it that only Bush 41 and Bush 43 voter turnout had to exceed 100% (based on Final National Exit Polls in 1988, 1992, 2004 and 2008) in order to match the recorded vote? AND YET OTOH HAS NOT MENTIONED THAT UNCOUNTED, SWITCHED OR STUFFED VOTES COULD BE THE REASON FOR THESE RECURRING ANOMALIES.

OTOH:
I am aware of several European sources that describe vote misreporting over an extended period. Noelle-Neumann’s The Spiral of Silence (1984/1993) notes that reported voting for a particular party tends to covary with the party’s current popularity. In Britain, Himmelweit et al. (1978) reported that in a longitudinal study, 16% or more of participants misrecalled their prior votes four to eight years later, and that this difference tended to favor “the party for whom the subject had just voted” (369). The authors argued that “Political scientists have tended to overestimate consistency in electoral behaviour…. The voter also tends to overestimate his own consistency; so there is a conspiracy of error, of which political scientists in looking for trends need to be aware” (1978, 374). MORI (2001) presents an extensive time-series of recalled vote in the British 1997 General Election, in which Labour’s margin over the Conservatives is overstated by 10 points in the month after the election, and by as much as 17 points in subsequent surveys – a phenomenon they dub “inaccurate recall.” MORI uses these results to argue against the use of recalled past vote in weighting; two ICM researchers concede the existence of “faulty memory” or “false memory” while defending their weighting practice (Curtice and Sparrow 2002), which incorporates an adjustment for false memory. There is little reason to expect the dynamics of U.S. presidential elections to match those of parliamentary elections in Germany or GreatBritain.

In the U.S. context, an immense literature discusses the phenomenon of vote overreporting, especially in connection with the American National Election Studies’ vote validation studies. Cassel and Sigelman (2001) cite 28 such studies. Traugott (1989, 5) notes that in five validation studies from 1964 through 1988, “we find that 12-14% of self-reported voters do not actually have a record of voting.” Burden (2000) notes that the gap between NES and official estimates of voting-age turnout more than doubled from 1952 to 1996. The demographic correlates of overreporting –most of them modest – have been extensively considered, as have possible impacts on models of turnout and candidate choice. These studies rarely describe the reported voting behavior of the apparent false reporters, perhaps in part because the data are so sparse.12 following question asked if the respondent had voted; I cannot now determine what proportion of the self-reported voters implied that they had actually voted for Kennedy.)

TIA:
The fact that a percentage misreport is not the issue here. Lindeman provides no RATIONALE for his claim that a preponderance of Gore voters over Bush voters misreported. The claim is highly dubious because a) Bush had a 48% approval rating and b) Gore was the popular (and electoral) vote winner in a stolen 2000 election.

Recall the original exit pollster reluctant Bush responder (rBr) theory that the massive exit poll discrepancy was due to Bush voters who shied away from the exit pollsters. But rBr was refuted by the exit pollsters themselves in the Final NEP by the 43/37 Bush/Gore weightings. They were left with the Hobson's choice: accept the impossible Final NEP or rBr. They could not have it both ways.

Here is another Hobson's choice: was the respondent "misreporting" due to a) the tendency of voters to ASSOCIATE with an unpopular incumbent with 48% approval (i.e. a latent "bandagon effect") or was it b) due to a "slow moving fog" causing Gore voters to actually FORGET that they voted for him - despite the Supreme Court decision stopping the recount in favor of Bush stole the election from Gore. The decision was a UNIQUE HISTORICAL ABOMINATION. This explanation flies in the face of common sense: one would expect nearly EQUAL misreporting on the part of Gore AND Bush voters.

Lindeman's argument relies on four tenuous assumptions:
1) The recorded vote is equal to the True vote. Obviously false.
2) Gore voters forgot and misreported; Bush voters remembered. Not plausible.
3) Bush 48% approval was not a factor. Not plausible.
4) The Final NEP (13,660 respondents) 43/37 Bush/Gore mix was based on a legitimate sample. But at 12:22am (13047 respondents) the split was 41/39. How could it have changed so abruptly? Let's take the exit pollsters' at their word: it is standard operating procedure to force the NEP to match the recorded vote.

Q.E.D.

PART II

TIA:
The NES study was to determine how they voted in the previous election. The exit pollsters asked how they just voted minutes before in the current election. Could the "slow-moving fog" have drifted in so quickly?

OTOH:
12 For what it is worth, of the five validation studies incorporated in the NES cumulative file (1964, 1976, 1980, 1984, and 1988), the false reporters reported voting for Jimmy Carter in significantly higher proportion than the validated voters in 1976 (65.5% vs. 48.6%, p < 0.001); any differences in the other four elections fall short of statistical significance. Cassel and Sigelman (2001) conclude that misreporting has little impact on candidate choice models in the three elections they examine, 1980 through 1988. (They do not examine 1976 because the vote validation procedures were improved in 1978 [2001, 645].)

TIA: Well, that is one report which seems to make sense intuitively.

OTOH:
6
Wright (1993) discusses both vote overreporting and the bandwagon effect. Wright (1993, 291) notes that in NES post-election interviews, a “significant number of respondents overstate their support for the winning candidates.” Importantly, this propensity is not limited to overreporters: “while those who cannot be shown to have voted are more easily moved to report going with the winner, they are followed rather closely by those who did vote” (1993, 304). However, Wright finds “no overall pattern” of pro-winner bias for presidential vote reports if the 1964 study – which grossly overstates Lyndon Johnson’s margin over Barry Goldwater – is excluded from the analysis (1993, 300). Here Wright expands upon earlier work exploring bandwagon or spiral-of silence effects favoring Senate winners (Wright 1990; Gronke 1992, in critique; Wright 1992, in response). Wright (1992, 132) suggested that given the high salience of the presidential vote, “although we may expect decay in recall of presidential voting, it may not occur appreciably for weeks or months after the election.”

TIA:
But Gore was the popular vote winner in 2000. And he won by much more than the official 540,000 margin. According to the Census, there were 5.4 million (net) uncounted votes of which Gore had approximately 4 million. Therefore it is safe to assume that he won by approximately 3 million votes.

OTOH:
Strangely, the bandwagon effect has been offered as further evidence that the 2004 election was stolen. Maxwell (2004) states: “It is a well-recorded phenomenon that after an election result is known, more people will claim to have voted for the winner than actually did…. In this case [2004], and as far as I can discover, only in this case does the percentage claiming to have voted for the winner fall below the percentage actually voting for him.” Maxwell cites Prisuta’s (1993) evidence of a presidential bandwagon effect in a telephone poll following the 1992 election, in early December. Among the respondents who reported having voted for one of the three leading candidates, Bill Clinton retrospectively held a 17-point margin over George H. W. Bush, 49% to 32% – whereas in the official returns Clinton won by only 5 points, 43% to 38%. Prisuta concluded that the largest disparities occurred in subgroups “where Clinton ran strongest and where Democrats traditionally do well” (2) – specifically, women, low income voters, and ethnic minorities. For instance, the ICR survey showed 43% of men voting for Clinton (compared with 41% in the exit poll), but 54% of women (versus just 45% in the exit poll).

TIA:
Apparently, then, 2004 is an exception: fewer claimed to have voted for Bush than the recorded vote indicates. That is congruent with the True Vote Model which indicates that Bush's 62 million recorded vote was inflated by 5-6 million (10%). Kerry won the True Vote by 67-57 million. Don't forget there were 3.4 million (net) uncounted votes.

OTOH:
However, judging from the Pew Research Center’s post-election polls for the past five presidential elections, Wright was right to downplay presidential bandwagon effects, notwithstanding the case of 1992 (which came too late for Wright to discuss) and the very different case of 1964. In Pew’s 2004 post-election survey, 49% of respondents reported having voted for Bush, 45% for Kerry, less than 1% for Nader, and 6% “Other/DK” (Pew Research Center 2004). Thus Bush held a 4-point retrospective margin, lightly larger than his 2.5-point margin in the official returns. Compare results for previous elections:

Table 1: winning margins in post-presidential-election Pew telephone surveys
Year; Pew margin for popular vote winner; Official popular vote margin; Difference




2004 4% Bush 2.5% +1.5%
2000 3% Gore)* 0.5% +2.5%
1996 6% Clinton 8.5% –2.5%
1992 13% Clinton 5.6% +7.4%
1988 10% Bush 7.7% +2.3%

* Bush held a tenuous electoral college lead at the time of the survey

TIA:
The Pew survey is compared to the recorded vote margin - not the True Vote.
Let's compare the True Vote margin to the survey:

Year True Margin (diff) Note
2004: Kerry by 7% (+11%) Questionable:Other/Nader/DK =7%? Recorded "Other" was 1%.
2000: Gore by 3% (0%) Exact match to Gore's True Vote
1996: Clinton by 10% (+4%) within 2% of Clinton's True Vote (within MoE)
1992: Clinton by 20% (+7%) within 3% of Clinton's True Vote (within MoE)
1988: Dukakis by 1% (+11%) Questionable

The 1992-2000 True Vote is within the MoE of the Pew survey;
1988 and 2004 are outside the MoE.



Pg 7

OTOH:
Note that in 2000, Gore narrowly won the popular vote, Bush was the presumptive leader (if not winner) in the electoral vote, and the outcome was undecided at the time of the poll. Regardless, among these Pew results, the bandwagon effect of 1992 is unusual, while 2004 does not stand out. The NES results, presented below, provide more support for bandwagon effects than indicated by Wright, but much less than suggested by Maxwell.

TIA:
Gore won the RECORDED vote by 540,000; he won the True Vote by at least 3 million based on total votes CAST.

OTOH:
The U.S. discussion of “bandwagon effects” appears more or less to end when the NES postelection wave does. Little if any discussion parallels the British consideration of recalled vote years in the future.

False past-vote reporting in the General Social Survey

As a baseline to explore the intertemporal dynamics of “false recall” in the U.S. presidential context, I present some vote-recall results from the General Social Survey since its inception in 1973. Of course, these results are by no means comparable to exit poll results, or even to NES results given the NES’s proximity to elections and emphasis upon political content. For simplicity, I present only the margin between the two major candidates; percentages for other candidates tend to vary much less over time. For each series, the first point at left reflects the winner’s initial margin in the popular vote (slightly negative for George W. Bush in 2000); the following points represent recalled vote margin in General Social Surveys.

Figure 2: recalled vote margins in General Social Surveys, 1973-2004
recalled vote, General Social Survey 1973-2004

Pg 8

(The results have been weighted by the number of adults in each household, to approximate an equal probability of selection across individuals rather than households, as suggested in the GSS Appendix A. Also, black oversamples have been filtered out. The sample sizes range from 852 to 1164 prior to the 1994 election, and from 1652 to 1993 thereafter. Thus, the nominal margins of error for the margin can be as large as ±7 points prior to 1994, and ±5 from 1994 on.)

Five of the eight recalled-vote series end distinctly above the actual vote margin, although Reagan’s “margin” in his second term (following the 1984 election) appears to stagnate until 1988. Two presidents end up roughly where they began: Richard Nixon’s recalled margin in 1977 roughly matches his actual margin in 1972 (after a distinct drop prior to his resignation), and Bill Clinton more or less breaks even over his first term (through 1998). One president loses ground: seven years after Reagan’s victory over Jimmy Carter, a plurality of respondents recalled having voted for Carter! These differences apparently reflect not only the relative popularity of the winners, but the name recognition and popularity of the losers. Carter and George H. W. Bush, the two presidents in this group who were defeated for reelection (Ford never having been elected), fare best. At the other extreme, Mike Dukakis – probably the most obscure of the losing candidates – falls from a single-digit margin in the 1988 election to a brutal 41-point retrospective loss in 1993.

TIA:
The MoE is too large to draw any conclusions. In 1972 Nixon won a landslide; the retrospective margin would reflect that. In 1980, Carter's True Vote was higher than the recorded vote (as it always is for the Democrat). As far as Dukakis' 41% retrospective loss, that says more about the survey sample design than it does about "false recall".

OTOH:
I eschew further analysis of the GSS data, because my main interest is in exploring the impact of false vote reports on election studies. Respondents may be more prone to false vote report in contexts where relatively few of the questions treat political issues, and when interviewed further from election day.

False past-vote reporting in the National Election Studies: an overview since 1948. In most National Election Studies from 1948 on, respondents have been asked in the pre-election survey whether they voted in the prior presidential election, and if so, for whom. Prior to 1964, respondents were asked about the major candidates by name only; from 1964 on, respondents have been prompted with both the names and the party affiliations of the major candidates.13 In the post-election survey, of course, they are asked whom, if anyone, they have just voted for. Thus, the NES offers at least two opportunities for false vote reporting: immediately after the election (or as immediate as the post-election wave), and four years later. (Given my specific interest in election dynamics, I omit results from the off-year NES studies.)

13 Strictly, in 1964, roughly half the respondents were asked if they had voted “Republican or Democrat for president in that [1960] election.” The results were not significantly different from the other half-sample. In that other 1964 half-sample, and in following years, the basic structure of the question (slightly modified for major third-party candidates) has been: “In <year>, {you remember that} <Republican presidential candidate> ran on the Republican ticket against <Democratic presidential candidate> for the Democrats. Do you remember for sure whether or not you voted for in that election? (IF YES, VOTED:) Which one did you vote for?”

The 1948 question also differed from the version asked in 1952 through 1960. In 1948, the wording was: “Do you remember whether you voted in <year> when <Democratic presidential candidate> ran against <Republican Presidential candidate> (IF YES:) Whom did you vote for then?”

In 1952 through 1960, the wording was, “In <year>, you remember that <Democratic presidential candidate> ran against <Republican presidential candidate>. Do you remember for sure whether or not you voted in that election? (IF YES:) Which one did you vote for?” Thus, the 1948 wording (which omits “for sure”) lends itself to more over-reporting.

TIA:
But again, the question is not who they said they voted for in the LAST election, but who the NEP respondents said they just voted for in the CURRENT ELECTION just MINUTES BEFORE.

Pg 9
OTOH:
For the comparison four years out, I offer two change measures. The first measure is limited to respondents who indicated that they had voted both four years ago and (in the post-election wave) that they had voted in the current election, and who named whom they had voted for in each. These results are presumably more nearly compatible with exit poll results. The second measure includes all respondents who reported a vote choice in the previous election, whether or not they reported voting in the current election. Table 2 summarizes the basic results.14

Table 2: recalled vote margins in National Election Studies, 1948-2004
Election year; 1-Winner’s official margin; 2-Winner’s margin; 3-NES post change in
margin; 4- NES post Winner’s margin; 5- NES +4 years* Change in margin; 6- NES +4 years(voters)* Change in margin; 7 NES +4 years (all)**




Year 1 2 3 4 5 6
1944 7.5% n/a n/a 20.6% 13.1% 22.4%
1948 4.5% 8.4% 3.9% 13.5% 9.0% 11.7%
1952 10.9% 16.1% 5.3% 26.3% 15.4% 16.7%
1956 15.4% 19.2% 3.8% 26.7% 11.3% 12.4%
1960 0.2% -1.7% -1.9% 27.7% 27.5% 27.6%
1964 22.6% 35.0% 12.4% 32.7% 10.1% 13.0%
1968 0.7% 6.7% 6.0% 18.9% 18.2% 13.0%
1972 23.2% 28.3% 5.2% 30.1% 7.0% 6.1%
1976 2.1% 2.3% 0.2% 2.8% 0.8% 5.6%
1980 9.7% 11.4% 1.7% n/a n/a n/a
1984 18.2% 16.3% -1.9% 25.5% 7.2% 6.8%
1988 7.7% 5.7% -2.0% 20.5% 12.8%16.0%
1992 5.6% 13.9% 8.3% 6.5% 1.0% 3.2%
1996 8.5% 15.5% 7.0% 14.9% 6.4% 12.7%
2000 -0.5% -4.2% -3.7% 9.1% 9.6% 5.6%
2004 2.5% 0.4% -2.1% n/a n/a n/a


* Pre-election NES four years later; limited to people who reported (in the post-election survey) having also voted in the later election.

** Pre-election NES four years later; all respondents to retrospective question
Although short-run bandwagon effects are not a major focus of my paper, they are salient because of the use I will later make of the 2000-2004 NES panel survey. If we had reason to believe that NES respondents in either 2000 or 2004 were radically prone to bandwagon effects, the panel results would be gravely (or more gravely) compromised. Earlier I reported Wright’s conclusion that apart from 1964, there was no evidence of a presidential bandwagon in the years he examined. Here, considering fourteen elections from 1948 on (excluding the problematic case of 2000 15), I do find some propensity for a bandwagon effect: the mean winning margin is 3.3

TIA: But...all the above retrospective statistics assume the RECORDED vote as a baseline for comparison and are therefore unrealistic. There is not even a mention of uncounted/stuffed votes. A proper analysis would be to add back the net uncounted votes to get a reasonable approximation of TOTAL VOTES CAST.

IN EVERY ELECTION, THE DEMOCRATIC TRUE VOTE MARGIN EXCEEDS THE RECORDED VOTE MARGIN SHOWN ABOVE. THAT'S BECAUSE THE VAST MAJORITY OF UNCOUNTED VOTES ARE DEMOCRATIC.

Comparing the True Vote margin (calculated for the 10 elections (1968-2004) to NES we find that 6 of the 10 NES results fall within a 4% MoE (700 sample).

So yes, we believe NES, but only if the responses are compared to the True Vote.




Winners recorded margin True margin; diff(ABS); share dev; within 4% MoE?
Year Rec NES Diff
1944 7.5% n/a n/a
1948 4.5% 8.4% 3.9%
1952 10.9% 16.1% 5.3%
1956 15.4% 19.2% 3.8%
1960 0.2% -1.7% -1.9%
1964 22.6% 35.0% 12.4%
True Diff Share
1968 0.7% 6.7% 6.0% -3.6% 10.3% 5.1%
1972 23.2% 28.3% 5.2% 16.9 11.4 5.7
1976 2.1% 2.3% 0.2% 6.8 4.5 2.3 Y
1980 9.7% 11.4% 1.7% 6.9 4.5 2.3 Y

1984 18.2% 16.3% -1.9% 16.4 0.1 0.05 Y
1988 7.7% 5.7% -2.0% -2.1 7.8 3.9 Y
1992 5.6% 13.9% 8.3% 21.0 7.1 3.5 Y
1996 8.5% 15.5% 7.0% 16.5 1.0 0.5 Y
2000 -0.5% -4.2% -3.7% 4.3 8.5 4.2

2004 2.5% 0.4% -2.1% N -8.1 8.5 4.2

Part III

OTOH:
Using the NES 1948-2000 cumulative file, I have applied the post-election
post-stratification weight VCF0009B; supplemented by the NES 2004 datafile,
again using the post-election post-stratification weight.

In 2000, Al Gore appears to increase his small lead in the popular vote; most of the
interviews were conducted before Gore conceded. In the other direction, a good case
could be made for excluding 1964, as Lyndon Johnson’s margin in the post-election
study (35 points) was substantially smaller than in the pre-election study (40 points) – hardly a prototypical bandwagon.

TIA:
Lindeman does not mention that in 2000, 5.4 million votes were uncounted
(180,000 in Florida alone), the vast majority Gore votes. He won the election by
at least 3 million votes.

OTOH:
Contrary to Pew, the NES indicates a bandwagon effect favoring Clinton in 1996.
However, in four of the last six elections, the NES margin is (insignificantly)
smaller than the actual margin.

The question wording of the vote question was substantially changed in 2000 (and
in a half-sample in 2004), reducing self-reported turnout rates; however, it is
unlikely that this reduction substantially affected estimates of bandwagon
effects.16

A long-term bandwagon effect – the apparent propensity to overstate having voted
for the incumbent four years earlier – is much more consistent in these data.
In all fourteen presidential elections for which the past-vote question was
available, the previous winner’s margin in the past-vote question was larger than
the official margin.

Among self-reported current voters, the average increase in retrospective margin
was 10.7 points. However, the effect appears to decline after 1964, when party
affiliation was first supplied in the retrospective vote question. From 1968 on,
the mean increase in margin is 8.1%.

The average increase among all respondents was 12.3 points; this increase was
larger than the increase among self-reported current voters in 10 of the 14 years,
arguably weakly supporting the inference that vote overreporters are more prone to
a bandwagon effect. (The bandwagon effect is positively correlated with presidential
approval, but the correlation falls well short of statistical significance.)

TIA:
Lindeman implicitly assumes ZERO fraud by implying the recorded vote count is the
same as the TRUE vote. We know that is not true. In fact, the NES responders were
telling the Truth.

The NES average Vote share deviated from the True Vote over the past 11 elections by
1.9%, well within the NES 4% MoE!




1968 1972 1976 1980 1984 1988 1992 1996 2000 2004 2008 Average

Recorded Share
Dem 42.9 36.2 50.1 41.0 40.6 45.6 43.0 49.2 48.4 48.3 52.9 45.3
Rep 43.6 60.7 48.0 50.7 58.8 53.4 37.4 40.7 47.9 50.7 45.6 48.9
Other 13.6 3.1 1.9 8.3 0.7 1.0 19.5 10.1 3.8 1.0 1.5 5.9

Margin -0.7 -24.4 2.1 -9.7 -18.2 -7.7 5.6 8.5 0.5 -2.5 7.2 -3.6

True Vote
Dem 45.3 40.2 52.6 42.7 43.0 50.6 51.9 53.7 50.3 53.4 58.1 49.2
Rep 41.9 57.0 45.7 49.6 56.4 48.4 30.8 37.2 46.1 45.3 40.5 45.4
Other 12.8 2.8 1.7 7.8 0.6 1.0 17.3 9.2 3.6 1.3 1.4 5.4

Margin 3.4 -16.9 6.8 -6.9 -13.4 2.1 21.0 16.5 4.3 8.1 17.6 3.9

True Vote
vs.Recorded
Margin 4.1 7.6 4.8 2.8 4.8 9.8 15.5 8.0 3.8 10.6 10.3 7.5

NES
Margin 6.7 -28.3 2.3 -11.4 -16.3 -5.7 13.9 15.5 4.2 -0.4 21.0 0.1

True Vote
vs. NES
Margin -3.3 11.4 4.5 4.5 2.9 7.8 7.1 1.0 0.1 8.5 -3.4 3.7

Dem 4.5 7.1 1.0 -3.4 2.3
Rep -3.3 11.4 4.5 2.9 7.8 0.1 8.5 4.6

True Vote
vs. NES
Share -1.6 5.7 2.3 2.2 1.5 3.9 3.6 0.5 0.0 4.2 -1.7 1.9
<4%MoE? Yes No Yes Yes Yes Yes Yes Yes Yes No Yes Yes



OTOH:
Change in margin clearly is not a pure measure of false past-vote reporting, because
it is affected by differential turnout. In any given election, voters for the
previous winner may vote at a higher (or lower) rate than voters for the previous
challenger. Unfortunately, the influence of false reporting itself swamps any effort
to measure differential turnout directly. In the 14 NES studies for which the past
vote question was available, 6 indicated a statistically significant difference in
turnout – in every case favoring the out-party candidate. Presumably this difference
largely reflects overreporting of voting for the previous winner among respondents
who subsequently report not voting in the current election.17 On average, reported
out-party turnout (i.e., turnout in the current election among people who claimed to
have voted for the major-party loser of the previous election) was 2.9 points higher
than reported in-party turnout.

TIA:
The following turnout sensitivity analysis shows that Kerry wins by approximately 7
million votes even assuming 98% Bush 2000 voter turnout and 90% Gore voter turnout.




Kerry Share
Bush Gore Turnout
Turnout 90% 92% 94% 96% 98%

90% 53.7% 54.0% 54.3% 54.6% 54.9%
92% 53.3% 53.6% 53.9% 54.2% 54.5%
94% 53.0% 53.3% 53.5% 53.8% 54.1%
96% 52.6% 52.9% 53.2% 53.5% 53.8%
98% 52.2% 52.5% 52.8% 53.1% 53.4%

Kerry Margin

90% 11.1 11.8 12.5 13.3 14.0
92% 10.1 10.9 11.6 12.3 13.0
94% 9.2 9.9 10.6 11.4 12.1
96% 8.2 8.9 9.7 10.4 11.1
98% 7.3 8.0 8.7 9.4 10.2


Bush Kerry Share
Turnout Gore %Uncounted
50% 60% 70% 75% 80%
90% 54.9% 54.9% 54.9% 54.9% 54.9%
92% 54.5% 54.5% 54.5% 54.5% 54.5%
94% 54.1% 54.1% 54.1% 54.1% 54.1%
96% 53.8% 53.8% 53.8% 53.8% 53.8%
98% 53.4% 53.4% 53.4% 53.4% 53.4%

Kerry Margin

90% 14.0 14.0 14.0 14.0 14.0
92% 13.0 13.0 13.0 13.0 13.0
94% 12.1 12.1 12.1 12.1 12.1
96% 11.1 11.1 11.1 11.1 11.1
98% 10.2 10.2 10.2 10.2 10.2



OTOH:
To focus on recent elections: the 2000 NES, which employed the more stringent
wording for current-election voting, indicates that 87.4% of Clinton 1996 voters
turned out again in 2000, compared to 95.3% of Dole 1996 voters. (Compare the exit
poll “evidence,” discussed earlier, that over 100% of Clinton voters turned out in
2000!) The 2004 NES indicates a more equitable repeat turnout: a statistically
indistinguishable 91.1% of Gore 2000 voters and 92.8% of Bush.

TIA:
Lindeman does not state that in 1996 approximately 9 million votes
(mostly Clinton's) were never counted.




Method: True Vote
2000 95% 96%
Cast Record Alive Voted Voter
1996 1996 1996 2000 2000 Turnout Mix Gore Bush Other Turnout/Alive
DNV - - - - 15.05 9.1% 52% 43% 5% -
Clinton 53.7% 56.34 47.40 53.53 51.39 51.39 48.7% 82% 15% 3% 96%
Dole 37.2% 39.03 39.20 37.08 35.60 35.60 33.8% 7% 91% 2% 96%
Perot 9.2% 9.64 9.68 9.16 8.79 8.79 8.3% 39% 49% 12% 96%

Total 105.02 96.28 99.77 95.78 110.83 Total 50.34% 46.06% 3.60% 4.28%
55.79 51.05 3.99 4.75

Recorded 48.38% 47.86% 3.76%
105.42 51.00 50.46 3.96



OTOH:
16 From 1952 through 1996, respondents were asked whether they had voted using
language similar to this (from 1996): “In talking to people about the election we
often find that a lot of people weren't able to vote because they weren't registered
or they were sick or they just didn't have time. How about you, did you vote in the
elections this November?” In 2000, a new wording was adopted. The preamble (the
first sentence) was unchanged, but then respondents were asked: “Which of the
following statements best describes you: One, I did not vote (in the election this
November); Two, I thought about voting this time - but didn't; Three, I usually
vote, but didn't this time; or Four, I am sure I voted?” In half-sample experiments
in the 2002 and 2004 NES, this new wording yielded self-reported turnout rates seven
to eight points lower (Martinez 2006, 4). However, as Wright (1993) observes,
apparent overreporters are only somewhat more likely to say they voted for the
winning candidate.

17 Indeed, the apparent difference in turnout between past in-party and out-party
voters is – by construction – highly correlated with the apparent difference in
bandwagon effect between self-reported current voters and all respondents, as
described in the preceding paragraph.

pg 11
2000 voters.

18 This result is arguably consistent either with the hypothesis that Bush 2000
voters actually turned out at a substantially higher rate than Gore 2000 voters (if
the out-party turnout was again overstated), or with the view that turnout was close
to equal. The 2000-2004 NES panel offers an alternative measure of repeat turnout in
which people’s 2000 vote is reported in 2000 rather than 2004.

19 By this (also problematic) measure, 95.5% of Gore 2000 voters and 96.2% of Bush
2000 voters voted again in 2004.20

False past-vote reporting in exit polls

I obtained all national presidential general election exit polls archived with
ICPSR: ten exit polls in eight elections dating back to 1976. Using the supplied
national weights, I computed the retrospective margin of the previous winner as a
proportion of those who reported having voted.

As with the NES, in every case, the winner’s retrospective margin in the exit poll
four years later was larger than his initial margin – approximately 11 points larger
on average. George W. Bush’s increase in 2004 recalling 2000 is somewhat below the
mean of these ten exit polls, and close to Bill Clinton’s increase in 2000 recalling
1996. The gap between exit poll and NES results appears to narrow from 1992 on,
perhaps due to changes in the exit poll wording. (The appendix presents these
changes over time.)

Table 3: recalled vote margins in national presidential exit polls, 1976-2004

Election year Winner’s; official margin; Winner’s margin; exit poll four years later;
Change in margin; official to exit poll +4; Change in margin; official
to NES +4



1972 23.2% (CBS) 27.1% 3.9% 7.0%
1976 2.1% (CBS) 13.0% 10.9% 0.8%
1980 9.7% (ABC) 21.9%(CBS) 22.0% 12.2% 12.3% n/a
1984 18.2% (ABC) 34.4% (CBS) 32.4% 16.2% 14.2% 7.2%
1988 7.7% (VRS) 28.4% 20.7% 12.8%
1992 5.6% (VNS) 9.2% 3.6% 1.0%
1996 8.5% (VNS) 16.4% 7.9% 6.4%
2000 -0.5% (NEP) 7.5% 8.0% 9.6%


These results certainly do not support the inference that the 8.0-point change in
retrospective margin in the 2004 exit poll evinces massive vote miscount favoring
Bush. But the question remains to what extent the boost may be influenced by
differential turnout rather than false recall. Advocates of the fraud hypothesis
have argued that 2004 turnout among Gore voters should have at least matched turnout
among Bush2K voters. These observers variously cite Gore voters’ anger over the 2000
denouement (although this anger was by no means universal), the intense 2004 Get Out
The Vote efforts (on both sides), and of course the high overall turnout, which is
often assumed to be inherently favorable to Democratic candidates. Pomper (2005),
following the weighted exit poll results, concludes on the contrary that turnout
among Bush2K voters was substantially higher. Campbell (2005, 237) observes that
states in which Bush did better had larger increases in turnout; this result might
conceivably hint that Bush2K voters turned out at a higher rate, but Campbell
ventures no such extravagant ecological inference. 21

18 This result combines half-sample results for two distinct wordings, as detailed
in footnote 16.
19 Thus the comparison is limited to respondents who (1) in 2000, reported having
voted for Bush or Gore and (2) also participated in 2004.
20 Here I apply the 2004 weight variable (i.e., a post-stratification weight that
attempts to compensate for panel attrition). The unweighted result is that 302 of
314 Gore voters (96.2%) and 304 of 315 Bush voters (96.5%) also voted in 2004.

Similarly, some analyses have found somewhat larger turnout increases from 2000 to
2004 in counties that favored Bush. In light of the exit poll finding (supported by
pre-election polls) that 2000 non-voters tended to favor Kerry in 2004, these
analyses suggest that Bush may have enjoyed at least some advantage due to
differential mobilization of Bush2K and Gore2K voters.

However, any such advantage seems unlikely to account for much of the increase in
retrospective margin.

“Retrospective defection” in the NES 2000-2004 panel study

To gain some insight about the possible dimensions and impact of false reporting in
the 2000 exit poll, I turn again to the NES 2000-2004 panel study. Clearly NES
results are not directly comparable to exit poll results, but they have one great
virtue: they allow us to see respondents actually changing their report of whether
they voted, and whom they voted for, in 2000. In the panel, 759 respondents reported
their 2000 presidential vote, or non-vote, in both 2000 and 2004.

TIA:
Once again, the NEP matched the True Vote to withion 1.9%, thus CONFIRMING election
fraud. The NES respondents did NOT misreport their past votes.

OTOH:
Note that in 2004, panel respondents were interviewed only after the election. Thus,
panel reports of 2000 vote are not directly comparable to reports from the main 2004
NES study, in which the question was asked in the pre-election wave. To give
context, I report both the (often very small) unweighted Ns and row percentages,
and the weighted percentages, omitting third-party voters for simplicity.

Table 4: reports of 2000 vote in 2000 and 2004, NES panel study
2004 report (unweighted) 2004 report (weighted) 2000 report No vote Gore Bush
No vote Gore Bush

No vote 68 (61.3%) 14 (12.6%) 26 (23.4%) 71.7% 6.6% 20.1%
Gore 7 (2.2%) 279 (88.9%) 23 (7.3%) 2.6% 86.5% 7.6%
Bush 5 (1.6%) 5 (1.6%) 303 (96.5%) 1.8% 1.3% 96.7%

Two results here are especially striking. First, people who reported in 2000 that
they had not voted are much more likely in 2004 to report having voted for Bush in
2000 than for Gore. Second, a substantively significant proportion of (2000
self-reported) Gore voters – over 7% – “retrospectively defect” to Bush by 2004,
while the rate of retrospective defection from Bush to Gore is much smaller.

21 Indeed, Campbell suggests that “high turnout may have favored the president
because marginal voters were more likely to be energized by the positive messages
of voting for Bush than the negative messages for voting against him” (2005, 237).

TIA:
High turnout (i.e. new voters) favored Kerry. That was obvious. And there were
approximately 22 million new voters.

pg 13

OTOH:
These changes in recall of whether and how respondents voted have substantial
effects on the estimated defection rates of Gore 2000 voters to Bush, and of Bush
2000 voters to Kerry. If one relies on the 2004 proportions (among panel respondents
who also answered the vote question in 2000, and who say they voted in 2004),
approximately 10% of Gore 2000 voters voted for Bush, and about 9% of Bush 2000
voters voted for Kerry – a virtual break-even. However, if one looks back to the
2000 vote reports, a different pattern emerges. Now (applying the 2004 panel
weights), approximately 14.1% of Gore 2000 voters voted for Bush, while about 6.7%
of Bush 2000 voters voted for Kerry! These proportions – derived from just 532 panel
respondents – are far from precise, but they defy assimilation to chance (chi-square p < 0.01).

TIA:
"Derived from 532 respondents"...as compared to 3,182 in the National Exit Poll.
IN ANY CASE, THE NES SURVEYS MATCHED THE TRUE VOTE TO WITHIN 1.9%!

OTOH:
Let us consider the potential pitfalls of extrapolation from the NES panel to
the exit poll. Most obviously, there is a disparity in sample frame: the NES is
not limited to actual voters. However, the evidence indicates that any (same-year)
“bandwagon effect” in the NES is minimal in general, and in 2004 in particular.
Moreover, vote overreporting in 2004 is smaller than in years prior to 2000, due
to the more stringent question wording. Thus, the inclusion of over-reporters may
not have much effect on the analysis. On the other hand, over-reporters may be more
prone to misreporting past votes, and so may lead to exaggerated retrospective defection rates.

TIA:
Finally, an admission and a caveat.

OTOH:
The fact that the retrospective vote question was asked after the election may tend
to increase both the retrospective bandwagon effect and the convergence between
recalled past and current vote. We have seen that in the 2004 NES, the recalled
Bush 2000 margin among self-reported 2004 voters was 9.1 points (a 9.6-point increase).

In the panel, the recalled margin was 13.9 points among self-reported voters using
the more stringent question wording (see footnote 16), and 23.2 points among
self-reported voters using the traditional wording. Much of this “bandwagon surplus”
under the traditional wording owes to people who had reported in 2000 that they did
not vote. Among these respondents in 2004, those who reported voting under the
traditional wording were more likely to retrospectively defect to Bush, and less
likely to retrospectively defect to Gore, than those who reported voting under the
more stringent wording.

Interestingly, the retrospective defection rates of Gore to Bush, and Bush to
Gore, both were higher under the more stringent wording than the more permissive
wording. Thus, it seems likely that respondents who self-reported not voting in
2000 are most prone to retrospective bandwagoning in a post-election interview
than in a pre-election interview, and that a more stringent current-vote filter
tends to mitigate the effect. Retrospective bandwagoning among respondents who
had reported voting for Gore or for Bush in 2000 may not be as subject to this effect.

Of course, the exit poll is not a pre-election interview! It seems plausible,
especially in the case where an incumbent is reelected, that any additional
retrospective bandwagoning between pre-election and post-election will owe more
to current vote choice than to some generalized desire to back whoever won –
and, therefore, that the exit polls will more closely match post-election than
pre-election results. Yet this inference is speculative at best, and our
speculations about 2000 non-voters appear especially prone to error.

TIA:
Back on the bandwagon effect? Bush had 48% approval.

OTOH:
NES respondents, and particularly panel respondents, are subject to panel
conditioning: the intensity of the interviews is likely to stimulate more thought at
least about the content of the questions, and therefore these respondents may be (in
this respect) less likely to misremember their votes four years earlier. Moreover,
the intensity of the NES interview may create a propensity to under-represent
non-politicals – although, given the countervailing intensity of the recruitment
efforts, it is hard to guess how this propensity might compare with the exit polls.

This propensity, as well as panel attrition, may partly account for the smaller
average bandwagon effects in NES studies as opposed to exit polls, although changes
in question wording greatly complicate any such comparison.

TIA:
This is jargon based on unproven psychological inferences. The scientific method
eschews conclusions based on beliefs sans data. Respondents are less likely to
misrepresent how they voted four years earlier rather than FIVE minutes ago? Not
likely!

OTOH:
Modeling retrospective defection in the 2004 exit polls: a baseline scenario
The exit poll and NES data are, in my judgment to date, too problematic (and, in
the NES case, too sparse) to lend themselves to high-powered statistical modeling.
Nonetheless, inquiring minds want to know the possible implications of the NES
results for estimates of differential turnout and defection rates. I therefore have
developed some frankly conjectural results based as closely as possible on the NES
panel results. Although I do not assert that the NES proportions are either
statistically robust or likely to be directly transferable, using them provides some
discipline to the exercise, since alternative assumptions seem ad hoc. Because my
model initially grew out of a controversy as to the accuracy of the 2004 vote count,
I accepted external stipulations (including critics’ assumptions) wherever possible.
The model, implemented as a spreadsheet, cascades through six steps in order to
explore how various assumptions about turnout, actual vote division by 2000 vote or
non-vote, and false report of the 2000 vote could reconcile the actual results with
the exit poll results.

TIA:
The 13047 respondents in the 1222am NEP is too sparse a sample taken immediately
after voting? And 70,000 is too sparse for the 50 state exit polls? I disagree.
But I agree that 600 NES respondents is too sparse.

OTOH:
For the baseline scenario, I somewhat arbitrarily decided to use entirely unweighted
exit poll results as my goal in step 6 (see the right side of Table 7 on page 17).
Here yet again, my criterion was partly rhetorical: the observations in the national
exit poll dataset “are what they are,” and no one can plausibly complain that they
are skewed to official election returns. As with a great deal of exit poll rhetoric,
this argument does not bear close scrutiny: surely the cases in the national sample
should be weighted to compensate for differential non-response and probability of
selection before being subjected to serious analysis. Edison/Mitofsky (2005, 20)
report that a national sample applying such weights (but not weighting to the
official results) would yield an estimated Kerry margin of about +3.0% rather than
+4.7%. These weights cannot be reconstructed from available data. However, we can
use the fully weighted data as an alternative goal for Step 6, as I do in Model 4 below.

TIA:
Finally, the DU “Game” thread.

OTOH:
Step 1: estimate the numbers of 2000 voters for Gore, Bush, and “other” who could
vote in 2004.

For sake of argument, I accept an estimate, offered on-line by the pseudonymous
“TruthIsAll,” that the potential 2004 repeat electorate can be reckoned as
96.52% of the 2000 electorate (based on an annual mortality estimate of
0.87%, multiplied by four years and subtracted from 100%). This estimate seems
not much worth refining, given the further arbitrary turnout assumptions.
I thus estimate approximately 49,229,000 Gore2K voters, 48,704,000 Bush2K voters,
and 3,815,000 ‘Other2K’ voters who could have voted in 2004.

TIA
A more appropriate rate is voter (age 18+) mortality - approximately 1.25% (5% over
4 years). The 0.87% rate is for total mortality. The higher voter mortality rate
means there were more new and fewer returning voters -which also means a bigger
Kerry margin.

OTOH
15
Step 2: based on “repeat turnout” estimates, allocate the 2004 (official) electorate
among Gore2K voters, Bush2K voters, Other2K voters, and DNV2Ks (who “did not vote”
in 2000).

As noted above, the literal NES panel evidence indicates very slightly higher 2004
turnout among Bush2K voters than Gore2K voters. However, again for rhetorical
purposes, I initially assume that Gore2K and Bush2K turnout were equal. I further
assume that Other2K turnout was slightly higher (the panel implies 100% turnout).
Unsurprisingly, the very high repeat turnouts implied by the panel do not allow for
sufficient “new” (Did-Not-Vote 2000) voters in the 2004 electorate. The ultimate
target is to have 18.4% of respondents state in the exit poll that they did not vote
in 2000. However, since the NES panel indicates that one quarter or more of people
who did not vote in 2000 will, four years later, report having voted, the actual
proportion of DNV2Ks must be higher.

By assumption, in my baseline scenario, repeat turnout is 91.0% among Gore2K voters,
91.0% among Bush2K voters, and 94.0% among Other2K voters; given these figures,
DNV2Ks comprise 24.2% of the 2004 electorate.

TIA
The True Vote model indicates that DNV2K comprised 22 million (16%) of the 2004
electorate)

OTOH
Step 3: make assumptions about how these groups actually voted, to approximate the
actual vote.

These proportions are influenced by the observed exit poll and NES panel estimates,
but as with the Did Not Vote proportion in Step 2, the actual proportions here are
not expected to match the observed exit poll proportions, because the exit poll
does not measure actual 2000 vote.

The assumptions at Step 3 are also crucial “results” of the model. In the baseline
scenario, these values are estimated (using Excel’s Solver) so as to approximate
the official vote totals and to minimize the sum of squared deviations, in Step 6,
between the twelve modeled exit poll results and the actual (unweighted) results.22

TIA
"The exit poll does not measure the actual 2000 vote". Well, the True Vote model
does. It uses the 2000 recorded vote adjusted for 5.4 million net uncounted votes
and the 2004 recorded vote adjusted for 3.4 million. In other words, it uses TOTAL
VOTES CAST. Including (allocated) uncounted votes is absolutely necessary in order
to determine the True Vote. If one believes that the recorded vote is sacrosanct,
one must also believe that uncounted votes are irrelevant. Might as well ignore
voter mortality and a feasible estimate of voter turnout.

OTOH
Table 5: 2004 by 2000 vote proportions, baseline scenario assumed
Vote Turnout Kerry Bush NEP
Gore 91% 85.4% 14.1% 91%
Bush 91% 7.6% 91.9% 10%
Other 94% 65.1% 18.5% 64%
DNV n/a 50.8% 48.5% 57%

These percentages are all plausibly close to NES results. Note that the defection
rate of Gore2K voters to Bush (in 2004) approaches twice the defection rate of
Bush2K voters to Kerry (actually a smaller difference than in the NES panel).

TIA:
The vote shares are IMPLAUSIBLE. Just calculate the MoE for each category and add
a 30% cluster factor. The probability that vote shares would deviate as indicated
from the 1222am NEP is effectively ZERO. Again, to assume a 700-respondent
retrospective NES survey to a 13047 respondent NEP survey taken just minutes after
voting is embracing a faith-based analysis.

OTOH:
22 More specifically, the sum of absolute deviations between modeled and official
vote counts (Bush, Kerry, and Other) was capped at 50,000 votes – a value that
allows a maximum deviation of 25,000 votes or about 0.02% of total vote for any
figure. Also, the sum of Kerry and Bush percentages is capped at 99.5% in each case.

Pg 16
Step 4: apply assumptions about differential response rates to yield a Kerry margin
of approximately +4.7%, as in the unweighted national exit poll.

I set the Kerry participation rate per million at 120.0 and the Bush rate at 103.9,
implying an overall participation bias “alpha” of about 1.155 (cf. Liddle 2005;
Lindeman, Liddle and Brady 2005). The actual extent of participation bias may well
have been quite different among different groups of voters; it clearly varied from
state to state, and in other ways poorly documented or understood. Thus, the
assumption of a fixed bias may well distort other estimates. Yet no alternative
assumption has clear warrant.

TIA
Assuming that the Kerry participation rate is much higher than Bush is resurrecting
the totally debunked rBr hypothesis and by itself invalidates the model immediately.
It's one thing to make assumptions which are based on historical data: undecided
voter allocation, uncounted votes, mortality, voter turnout, etc. It is quite
another to assume what has already been thoroughty debunked.

OTOH
Step 5: set assumptions about “retrospective defection” rates (2000 actual vote to
2000 recalled vote), and the marginal vote rates for Bush and Kerry in 2004 among
each group of defectors.

In the baseline scenario, most of these values are taken or slightly rounded from
the NES 2000-04 panel, using unweighted results. However, one large adjustment is
necessary, because the panel does not incorporate young 2004 voters who were not
eligible to vote in 2000. I assume that 26% of DNVs were these new young voters.23
Because the youngest cohort of DNVs in the exit poll (those aged 18 to 24) appears
to have divided its votes fairly equally between Bush and Kerry (giving Kerry about
a 2-point margin in the unweighted results), I do not treat them as a separate
voting bloc in Step 3, but their inclusion does entail that the defection rate among
all DNVs is lower than it would be among DNVs who participated in the panel.

TIA:
Which exit poll time line is Lindeman referring to? At 12:22am, the 18-29 group
voted for Kerry by 56-42. Note the abrupt changes in the weights and vote shares
between 12:22am and the Final - for just 613 additional respondents.

http://www.geocities.com/electionmodel/NEP13047.mht

All numbers are percentages

[code]

Respondents
4pm 730pm 1222am Final 4pm 730pm 1222am Final
Age 8349 11027 13047 13660 8349 11027 13047 13660
Weights Kerry vote share
18-29 15 17 17 17 56 56 56 54 <<
30-44 27 27 27 29<< 48 49 49 46 <<<
45-59 31 30 30 30 52 51 51 48 <<<
60+ 27 26 26 24<< 48 48 48 46 <<

Total 100 100 100 100 50.44 50.53 50.53 47.96

VOTE (in millions) 61.67 61.78 61.78 58.64



OTOH:
I model seven groups of “retrospective defectors” that are large enough to be
somewhat measurable, although several of these obviously are inconsequential. Again,
the proportions are based on observed counts of (1) 2000 report of 2000 vote, (2)
2004 report of 2000 vote (yielding retrospective defection rates), and (3) 2004
report of 2004 vote (yielding marginal vote rates).

Table 6: retrospective defection and vote rates, baseline scenario
Retrospective defection type; Defection rate (% of group as per 2000 report);
% of defectors voting for Kerry; % of defectors voting for Bush

Gore Bush 7.3% 26% 65%
Bush Gore 1.6% 79% 21%
DNV Gore 9.4% 95% 5%
DNV Bush 17.5% 25% 75%
DNV Other 1.5% 60% 15%
Gore DNV 2.2% 57% 14%
Bush DNV 1.6% 20% 20%


23 The actual figure is probably lower. In the exit poll, about 35% of respondents
who reported not having voted in 2000 were in the 18-24 age cohort. Thus – if vote
overreporting does not vary sharply with age -- perhaps on the order of 20% were
actually ineligible to vote based on age. The effect of overstating this proportion
is to reduce the estimated DNV retrospective defection rates.

17
Step 6: compare the resulting expected exit poll counts of Kerry and Bush interviews
attributed to various 2000 vote statuses to the observed unweighted exit poll
results.

For the baseline scenario, the assumptions in Steps 1 through 5 yield expected exit
poll results that vary from the observed results by fractions of one point. (The sum
of squared deviations across the twelve parameters is approximately 1.96x10-4, a
root mean square of 0.40%.) The modeled results give a larger surplus of “Bush2K”
voters over “Gore2K” voters, offset by a larger net propensity of “Bush2K” voters to
defect to Kerry.

Table 7: Modeled exit poll results, baseline scenario
Modeled exit poll results; Actual exit poll results ;Reported 2000 vote;
% of voters %Kerry %Bush %Of voters %Kerry %Bush

Gore 38.3% 91.5% 8.0% 38.4% 90.9% 8.4%
Bush 40.2% 10.5% 89.1% 39.5% 9.8% 89.4%
Other 3.3% 69.3% 17.1% 3.7% 69.2% 17.1%
DNV 18.3% 56.6% 42.6% 18.4% 56.2% 42.6%


Perhaps the most striking aspects of the baseline scenario are the (arbitrary)
assumption of equal repeat turnout rates among Bush2K and Gore2K voters; the
asymmetrical defection rates, where 14.1% of Gore2K voters defect to Bush as
opposed to 7.6% of Bush2K voters to Kerry; and Kerry’s narrow margin (2.3 points)
among DNV2Ks, at odds with the 13-point gap in the weighted reported results.

TIA
This is the sensitivity analysis that Lindeman should have calculated:




Kerry Share

Share of Share of DNV
Bush 55% 56% 57% 58% 59%

12% 53.9% 54.0% 54.2% 54.3% 54.5%
11% 53.5% 53.6% 53.8% 53.9% 54.1%
10% 53.1% 53.2% 53.4% 53.5% 53.7%
9% 52.7% 52.8% 53.0% 53.2% 53.3%
8% 52.3% 52.5% 52.6% 52.8% 52.9%

Kerry Margin

12% 10.8 11.0 11.2 11.3 11.5
11% 10.3 10.5 10.7 10.9 11.1
10% 9.8 10.0 10.2 10.4 10.6
9% 9.3 9.5 9.7 9.9 10.1
8% 8.8 9.0 9.2 9.4 9.6


Kerry Share

Mortality Gore Turnout
96% 97% 98% 99% 100%

6.0% 53.14% 53.28% 53.43% 53.57% 53.72%
5.5% 53.11% 53.26% 53.41% 53.55% 53.70%
5.0% 53.09% 53.24% 53.39% 53.54% 53.68%
4.5% 53.07% 53.22% 53.37% 53.52% 53.67%
4.0% 53.05% 53.20% 53.35% 53.50% 53.65%

Kerry Margin

6.0% 9.6 9.9 10.3 10.6 11.0
5.5% 9.5 9.9 10.2 10.6 11.0
5.0% 9.4 9.8 10.2 10.5 10.9
4.5% 9.4 9.8 10.1 10.5 10.9
4.0% 9.3 9.7 10.1 10.4 10.8



OTOH

Alternative models

Alternative assumptions provide further insight into the behavior of the model.
For Model 2, I preserved the assumptions that the unweighted NES panel yields best
estimates of retrospective defection and voting rates (Step 5) and that the
unweighted exit poll results are an appropriate target (Step 6).

I freed the repeat turnout coefficients (Step 2), except that I imposed a ceiling
of 98% repeat turnout among Other2K voters – a ceiling that the model in fact hit.
The resulting model has substantially better fit than the baseline model, with a
root mean square of 0.17% compared to 0.40% in the baseline. Perhaps
counterintuitively (and very possibly incorrectly), this new model yields a higher
repeat turnout rate among Gore2K than Bush2K voters – 91.8% versus 89.3% – because
in the baseline scenario, self-reported “Bush2K” voters were somewhat
overrepresented. The defection rates diverge somewhat (14.6% of Gore2K voters to
Bush; 7.0% of Bush2K voters to Kerry) to offset the relative increase in Gore2K turnout.

Table 8: 2004 by 2000 vote proportions, Model 2 (unequal turnout)
Assumed past vote Turnout; Kerry %; Bush %

Gore 91.8% 84.9% 14.6%
Bush 89.3% 7.0% 92.5%
Other 98.0% 65.2% 18.6%
DNV n/a 50.7% 48.7%


Pg 18
Table 9: Modeled exit poll results, Model 2 (unequal turnout)
Modeled exit poll results; Actual exit poll results Reported 2000 vote % of voters
% Kerry % Bush % of voters % Kerry % Bush

Gore 38.6% 91.1% 8.5% 38.4% 90.9% 8.4%
Bush 39.6% 10.0% 89.6% 39.5% 9.8% 89.4%
Other 3.4% 69.3% 17.1% 3.7% 69.2% 17.1%
DNV 18.5% 56.4% 42.8% 18.4% 56.2% 42.6%


In Model 3 (details not reported), I used retrospective defection percentages
(Step5) based on weighted, instead of unweighted, results from the NES panel.
The fit of this model is worse than Model 2, but better than the baseline
(root mean square= 0.36%); the model generates too many self-reported DNV
voters because the “defection” rates are smaller. In this model, Gore2K turnout
is 2.9 points higher than Bush2K turnout, and the defection rates are estimated
at 14.2% Gore2K; Bush and 7.0% Bush2K&;Kerry – results similar to Model 2.

In Model 4, I returned to the unweighted retrospective defection percentages for
Step 5, but I used weighted exit poll results for Step 6 (and accordingly set the
participation ratio in Step 4 to 1.

TIA
This is a very thorough analysis which is very flawed. It was stillborn when the
assumptions were tied to the rBr hypotesis - and compounded when the NES was
compared to the recorded vote and not the TRUE vote. Rather than utilize factual
data and plausible assumptions, this exposition is just a Bridge to Nowhere.
Lindeman does not prove his case.

The True Vote model indicates that Kerry won BEYOND A REASONABLE DOUBT with an
analysis based on historic voting data and a set of feasible assumptions. The
sensitivity analysis diplays the effects of changes in model assumptions and other
parameters on the total vote shares.

Lindeman is obviously an inexperienced modeler. His smooth verbiage smothers the
bogus assumptions he uses in his model in a futile attempt to promote "false recall".

Part IV

OTOH
In Model 4, I returned to the unweighted retrospective defection percentages for Step 5, but I used weighted exit poll results for Step 6 (and accordingly set the participation ratio in Step 4 to 1.0, indicating no bias). These weighted results should be easier to fit to, and they are. The resulting root mean square is just 0.11%, indicating a very close match, as is evident below.

Table 10: 2004 by 2000 vote proportions, Model 4 (weighted exits)Assumed



2000 Turnout Kerry Bush
Gore 92.3% 85.3% 14.2%
Bush 94.0% 7.5% 91.9%
Other 98.0% 68.9% 20.2%
DNV n/a 52.1% 46.8%

Table 11: exit poll results, Model 4 (weighted exits)
Modeled exit poll results Actual (wgtd.) exit poll results Reported
2000 vote % of voters % Kerry % Bush % of voters % Kerry % Bush
Gore 36.8% 89.7% 9.8% 36.7% 89.7% 9.9%
Bush 43.0% 9.1% 90.4% 43.0% 8.8% 90.2%
Other 3.3% 69.7% 20.2% 3.4% 69.7% 20.2%
DNV 16.9% 53.5% 45.0% 16.9% 53.4% 44.9%



In Model 4, repeat turnout among Bush2K voter narrowly exceeds turnout among Gore2K
voters, more consistent with the NES results. In fact, Bush2K voters outnumber Gore2K voters in the actual electorate, but only very narrowly so (37.4% versus 37.2%). The defection rates are similar to the baseline scenario, but new (DNV 2000) voters give Kerry a respectable 7-point margin, still substantially smaller than in the weighted exit poll results.

Model 5’s assumptions are identical with Model 4’s except that the retrospective defection rate of Gore2K voters to “Bush2K” voters is reduced from 7.3% (see Table 6) to 6.0%. This model yields a similar degree of fit (RMS = 0.12%, details not reported). The gap in defection rates narrows, while the turnout gap widens (the Bush turnout actually reaches its upper constraint at 95%). Bush2K voters now outnumber Gore2K voters in the 2004 electorate by 1.2 points (37.8% to 36.6%).

Table 12: 2004 by 2000 vote proportion, Model 5 (weighted exits, reduce false report)
Assumed


2000 Turnout Kerry Bush
Gore 91.0% 86.0% 13.5%
Bush 95.0% 7.7% 91.6%
Other 98.0% 68.9% 20.2%
DNV n/a 52.1% 46.8%


Of course, alternative models can be multiplied indefinitely. For instance, if the retrospective defection rates among DNV2Ks (people who initially reported not having voted in 2000) to both “Bush2K” and “Gore2K” are increased by 20% (not 20 points) each, to 21.0% and 11.3% respectively, the Gore→Bush defection percentage increases to 14.2%, the Bush→Kerry defection percentage decreases to 6.9%, and the turnout gap narrows (repeat turnout 91.2% for Bush2K, 90.6% for Gore2K). Under a range of plausible assumptions, it can be said that the Gore defection percentage is “around 14%,” the Bush defection percentage is “around 7%,”and the Gore2K and Bush2K shares of the 2004 electorate are “similar.”


OTOH
Conclusion

False vote recall complicates our analysis of partisan dynamics, and challenges some unconscious assumptions. Political observers rarely profess surprise that some respondents wrongly claim to have voted, but some find it strange that millions of voters might misreport – indeed, might forget – whom they voted for four years ago. I have not presented (or uncovered) systematic evidence about the mechanism behind false vote recall, but mere forgetfulness is not a bad account for respondents who (e.g.) report in 2000 that they voted for Gore, then report four years later that they had voted for Bush in 2000 but for Kerry in 2004.

No spiral of silence this: more like a slow-drifting fog. I am reminded of Larry Bartels’ (1996) conclusion that presidential incumbents derive approximately a five-point advantage from “information effects” (or, one might say, non-information effects) in the electorate. False vote recall favoring the previous winner is one distinctive manifestation of this incumbency advantage, although its practical importance is difficult to gauge – especially given the confounding influence of differential turnout.

George W. Bush evidently won in 2004 not by turning out a higher proportion of his 2000 supporters, but (inter alia) by winning the votes of millions of people whom, if asked, would not have recalled that they did not vote for him the first time around. The fraud theorists were right to infer that the previous-election tabulation could not mean what it said, and their account of it – a desperate attempt to paper over the evidence of a stolen election – has evident narrative appeal. On the evidence presented here, however, retrospective Bush bandwagoning is what we should have expected all along.

TIA:
Conclusion

No, what we should have expected all along was that the media would go to any extreme to cover up the 2004 election theft. Otherwise, the endemic fraud which has permeated our elections for many years might also be uncovered.

False Recall was advanced as soon as the reluctant Bush responder (rBr) theory was thoroughly debunked. Exit poll naysayers could not provide a rational explanation for the Final 2004 National Exit Poll and its impossible returning Bush phantom voter anomaly. They had to come up with another explanation. But the problem, as Lindeman himself admits, is that even assuming that a small percentage of voters forget their past vote, there is no rational evidence to explain why this phenomenon would invalidate the exit polls.

Consider this. Why was "mere forgetfulness" unique to Gore or Kerry voters? Are we supposed to believe that only those who DID NOT vote for an ILLEGITIMATE incumbent were afflicted with Alzheimer's? How come returning Bush voters were immune to this malady?

The basic premise for "false recall" is based on the NES limited sample survey which implies that the recorded vote is the same as the True Vote. Obviously it is not; there are millions of uncounted votes in every election. That is a FACT. And we know that votes were miscounted (see Ohio) and ballots stuffed. Uncounted votes alone should obviate the need to further debunk "false recall".

Nonetheless, I have presented a True Vote analysis which should finally put "false recall" to rest. Remember, the original purpose for proposing it was to call into question the 2004 exit poll results. The MSM fiction that Bush really did win the election fairly had to be maintained. Therefore, the recorded vote was sacrosanct; the corollary was that the unadjusted (unweighted) exit polls were not accurate (after all they differed from the recorded vote) and could not be cited as indicators of fraud. So much for the spreadsheet-wielding Internet bloggers.

The "confounding influence of differential turnout" is just fancy jargon to impress the unwashed. As I demonstrated above, a robust sensitivity analysis of varying Gore and Bush voter turnout (based on a solid foundation of relevant data) indicates that Kerry wins BY 7 MILLION VOTES - EVEN ASSUMING 98/90% BUSH/GORE DIFFERENTIAL TURNOUT. DIFFERENTIAL TURNOUT IS NOT A CONFOUNDING VARIABLE AFTER ALL.

The sole basis for Lindeman's case is a retrospective NES study of 700 respondents who were asked about their past vote. But there were also 51 pre-election state polls (30,000 respondents) and 18 national polls (27,000 respondents) taken a few DAYS BEFORE THE ELECTION and 51 post-election state (70,000 respondents) and the National Exit Poll (13,047 respondents) TAKEN IMMEDIATELY AFTER VOTING. So a survey of respondents taken FOUR years AFTER an election renders pre-election polls taken JUST BEFORE the election, and exit polls taken JUST MINUTES AFTER VOTING, OBSOLETE?

It should be obvious to any impartial observer that there is no basis for comparison between the limited NES samples and the massive pre-election and exit poll samples. The evidence is overwhelming and BEYOND A REASONABLE DOUBT that Kerry won the election. The only question is by how much.

- The True Vote Model indicates that Kerry won by 53-46% - a 10 million vote margin.
- The final 18 national pre-election polls (after undecided voters were allocated) projected that Kerry would win by 51-48%.
- The unadjusted state exit poll aggregate (WPE/IMS) indicated he won by 52-47%.
- The adjusted state exit poll aggregate ("Best GEO") indicated it was 51-48%.
- The adjusted ("Composite") state exit poll (12:40am) indicated it was 50.5-48.5%.

The Final National Exit Poll was FORCED TO MATCH THE RECORDED VOTE USING AN IMPOSSIBLE NUMBER OF RETURNING BUSH 2000 VOTERS.

The impossible Final NEP confirmed the impossible, fraudulent recorded vote:
Bush by 50.7-48.3%.

---------------------------------------------------------------------

OTOH
The comments, attributed to “josueencuentro” and “Robert Miller” respectively, responded to post “Moral Values”(11/7/04), at http://www.mysterypollster.com/main/2004/11/moralvalues.html

http://www.jamaicaobserver.com/columns/html/20041106T200000-

See Freeman (2005a); Freeman (2005b); also Lohrentz 2005, Keefer 2005, “TruthIsAll” 2005.0500_68984_OBS_A_LOBOTOMY_FOR_DEMOCRACY.asp (last accessed 3/19/06).
MORI. 2001. “How did you vote?” Research review, 3 April 2001. Published on the Internet,
http://www.mori.com/mrr/2001/c010403.shtml (last accessed 4/10/06).

Noelle-Neumann, Elisabeth. 1984/1993. The Spiral of Silence: Public Opinion – Our SocialSkin. Second edition, 1993. Chicago: University of Chicago Press.
Pew Research Center for the People and the Press. 2004. “Voters Liked Campaign 2004, But

Twoo Much ‘Mud-Slinging’: Questionnaire.” Released November 11, 2004 (field dates 11/5-8). Published on the Internet, http://people-press.org/reports/print.php3?PageID=912 (last accessed 3/19/06).

Pomper, Gerald M. “The Presidential Election: The Ills of American Politics After 9/11.” In Michael Nelson (ed.), The Elections of 2004 (Washington, DC: CQ Press, 2005), pages 42-68.

Prisuta, Robert H. 1993. “A Post-Election Bandwagon Effect? Comparing National Exit Poll

Data with a General Population Survey.” Proceedings of the Survey Research Methods Section, American Statistical Association.
http://www.amstat.org/sections/SRMS/Proceedings/papers/1993_194.pdf (last accessed 3/18/06).

Traugott, Santa. 1989. “Validating Self-Reported Vote: 1964-1988.” Prepared for the Annual Meeting of the American Statistical Association, August 7-10, 1989. NES Technical Report Series, No. nes010152. Available on the Internet,
ftp://ftp.nes.isr.umich.edu/ftp/nes/bibliography/documents/nes010152.pdf
(last accessed 4/10/06).

“TruthIsAll.” 2005. “Naysayer Hobson’s Choice: Final NEP or rBr? Take your pick.” Posted on the Internet, June 23, 2005, on DemocraticUnderground.com,
http://www.democraticunderground.com/discuss/duboard.php?az=show_mesg&forum=203&top
ic_id=380085&mesg_id=380085
(last accessed 3/18/06).

Wright, Gerald C. 1990. “Misreports of Vote Choice in the 1988 NES Senate Election Study.” Legislative Studies Quarterly 15, 4 (November 1990): 543-563.

Wright, Gerald C. 1992. “Reported versus Actual Vote: There Is a Difference and It Matters.” Legislative Studies Quarterly 17, 1 (February 1992): 131-142.

Wright, Gerald C. 1993. “Errors in Measuring Vote Choice in the National Election Studies, 1952-88.” American Journal of Political Science 37, 1 (February 1993): 291-316.

Phantom Bush Voters: True Vote Recall and the 1968-2008 Elections

TruthIsAll

August 9, 2009

http://www.geocities.com/electionmodel/FalseRecallRebuttal.htm

This is the conclusion from "Too Many Bush Voters? False Vote Recall and the 2004 Exit Polls" by Mark Lindeman
http://inside.bard.edu/~lindeman/too-many.pdf

“False vote recall complicates our analysis of partisan dynamics, and challenges some unconscious assumptions. Political observers rarely profess surprise that some respondents wrongly claim to have voted, but some find it strange that millions of voters might misreport – indeed, might forget – whom they voted for four years ago. I have not presented (or uncovered) systematic evidence about the mechanism behind false vote recall, but mere forgetfulness is not a bad account for respondents who (e.g.) report in 2000 that they voted for Gore, then report four years later that they had voted for Bush in 2000 but for Kerry in 2004.

No spiral of silence this: more like a slow-drifting fog. I am reminded of Larry Bartels’ (1996) conclusion that presidential incumbents derive approximately a five-point advantage from “information effects” (or, one might say, non-information effects) in the electorate. False vote recall favoring the previous winner is one distinctive manifestation of this incumbency advantage, although its practical importance is difficult to gauge – especially given the confounding influence of differential turnout.

George W. Bush evidently won in 2004 not by turning out a higher proportion of his 2000 supporters, but (inter alia) by winning the votes of millions of people whom, if asked, would not have recalled that they did not vote for him the first time around. The fraud theorists were right to infer that the previous-election tabulation could not mean what it said, and their account of it – a desperate attempt to paper over the evidence of a stolen election – has evident narrative appeal. On the evidence presented here, however, retrospective Bush bandwagoning is what we should have expected all along”.

No, what we should have expected all along was that the media would go to any extreme to cover up the 2004 election theft. Otherwise, endemic fraud that has permeated our elections for many years might also be uncovered.

False recall was based on an NES 600-sample survey. It compared the respondent’s recall of their past vote to the recorded vote. But there are millions of uncounted votes in every election. And we know that votes were miscounted and ballots stuffed (see Florida 2000 and Ohio 2004). By not considering total votes cast, “false recall” is based on the false premise that the recorded vote is identical to the True vote and that elections are fraud-free.

False Recall was advanced immediately after the reluctant Bush responder (rBr) theory was thoroughly debunked. The exit poll naysayers could not provide a rational explanation for the Final 2004 National Exit Poll’s impossible returning Bush voter anomaly; they had to come up with another explanation. It was the equivalent of a “Hail Mary” pass; there was no other way to rebut the analysis of countless researchers that the 2004 exit polls provide overwhelming evidence that the election was stolen. Even assuming the unlikely scenario that a significant percentage of voters forgot their past vote, that does not explain why Gore voters would forget or misreport their vote at a much higher rate than Bush voters. They could not offer a plausible rationale for “false recall”.

False recall was proposed to call into question the 2004 unadjusted and preliminary exit poll results. The media fiction was that Bush won the election fairly; it had to be maintained that the recorded vote sacrosanct. Of course they must have also considered that the discrepancy was due to massive fraud – but never dared to mention it. Therefore, they had to claim that the unadjusted (unweighted) exit polls were “bad” since they differed from the “actuals” - the official recorded vote.

The mathematically impossible Final National Exit Poll 43/37% returning Bush/Gore voter mix refuted rBr. The Hobson’s choice was to either accept a mathematically impossible Final Exit Poll or claim that exit poll responders misstated their 2000 vote due to “false recall”. They were forced to choose the latter.

But just 3200 of 13,000 NEP respondents were asked whom they voted for in 2000. What about the other 10,000 who weren’t asked the question? Surely a “slow-drifting fog” would not cause them to forget that they just voted for Kerry. So the only other explanation was that they intentionally misrepresented their vote. If so, what was their motive? Was it to jump on a four-year retrospective Bush bandwagon? What bandwagon? He had a 48% approval rating!

Why “false recall” was even suggested as an explanation is a mystery. Apparently, Lindeman never considered that NES would confirm the True Vote model - not refute it. His analysis was predicated on the premise that it was correct to use the recorded vote as a baseline (i.e. assume the election was fraud-free). In other words, a fraudulent recorded vote was assumed in concluding that the Final 2004 National Exit Poll 43/37 split was due to substantially more Gore than Bush voters misrepresenting their 2000 vote. That is a circular argument.

So which is it? Slow-drifting fog? Mere forgetfulness? Retrospective bandwagon?
The answer is: None of the above! The NES responders told the truth about their vote.

Comparing the average True Vote margin to the average NES margin for all 11 elections from 1968-2008 actually debunks “false recall”. The analysis confirms that election fraud is endemic; it always reduces the Democratic True vote! The average Democratic True Vote margin was 49.2-45.4%. In 8 of the 11 elections, the Democratic True Vote share fell within the NES 4% margin of error.

The average absolute NES/ True vote deviation for the 11 elections was -0.40%. Where’s the beef?

http://www.geocities.com/electionmodel/NESMargin.gif

The True Vote shares were within 1% of NES!

The average NES winning margin was 11.4% (11.4% for the Democrats and 11.5% for the Republicans).
The average Democratic True Vote winning margin was 10.0%
The average Republican True Vote winning margin was 12.4%

NES vs. True Vote Share (1968-2008)
The average absolute deviation for the 11 elections was -0.40%.
The average Democratic absolute deviation was -0.70%.
The average Republican deviation was 0.46%.

NES vs. Recorded Vote (1968-2008)
The average absolute vote share deviation was -1.75%.
The average Democratic absolute deviation was -3.30%.
The average Republican deviation was -0.46%.

http://www.geocities.com/electionmodel/NESMargin_TrueRecall.gif


The “confounding influence of differential turnout” is just fancy jargon to impress the unwashed. A robust sensitivity analysis (see below) of varying Gore and Bush voter turnout based on total votes cast indicates that Kerry would have won the True vote by 7 million - even assuming an implausible Bush/Gore voter differential turnout of 98/90% in 2004. He wins by 10.2 million assuming equal 98% turnout!

Differential turnout is not a confounding variable after all.

The exit poll debate was transformed into pseudo-psychological “false recall” conjecture. As a “spreadsheet-wielding Internet blogger” dealing with factual data, I never thought I would have to become an armchair psychologist. Let’s stipulate that humans tend to bury past transgressions into their unconscious.

It’s extremely unlikely that Gore voters would forget that the election was stolen from them in broad daylight; after all, they were the ones who were wronged. They would have no incentive, conscious or otherwise, to misrepresent their vote. Gore voters had nothing to be ashamed of. On the other hand, voting for Bush in 2000 was an act regretted by many (he had 48% approval on Election Day 2004). It’s more plausible that returning Bush voters would have regretted their past vote and tell the exit pollsters that they voted for Gore in 2000 – and that they just voted for Kerry.

Let’s not forget the 70,000 State exit poll respondents. Kerry won the unadjusted exit poll aggregate by 52-47% - a 6 million margin. Lindeman never discussed the state exit polls in his false recall argument. Rather, he wants us to believe that a 600 sample retrospective survey taken four years after the election renders null and void 18 national pre-election polls (27,000 total respondents); 50 state pre-election polls (40,000); 50 state exit polls (70,000) and the 12:22am National Exit Poll (13,047).

The simple fact is that in every election millions of votes are uncounted, therefore the recorded vote cannot represent the True Vote. The analysis presented by Lindeman implicitly assumes that the recorded vote is equal to the True Vote. But knowing the True Vote is what the exit poll debate is all about. Given the faulty premise, why even bother to continue to proceed further and refute his paper point by point?

The 2004 Census reported that 125.7 million votes were cast. Only 122.3 million were recorded; there were approximately 3.4 million net uncounted votes. Iin 1988, there were 10.6 million net uncounted votes. The Census margin of error for total votes cast is 0.30%.

The “Swing vs. Red-shift” argument was another canard meant to confuse and misdirect. The claim was that that fraud was unlikely since there was no direct correlation between swing (change in the Bush recorded vote from 2000 to 2004) and “red-shift” (the 2004 exit poll discrepancy). The argument was ridiculous on its face; a positive correlation is not a necessary requirement for fraud. Swing vs. Red-shift was formally debunked in a pure math logic proof by Kathy Dopp at US Count Votes.

Swing vs. Red-shift was based on the same invalid premise as “False Recall”: that the recorded vote was equivalent to the True Vote. Net uncounted votes were not factored into the equation even though uncounted and stuffed ballots caused the Democrats to lose millions of votes in every election. It was an exercise in circular logic by requiring that there must be a correlation between bogus recorded vote counts (“swing”) and exit poll (“red-shift”) discrepancies in order to prove election fraud.

A True Vote post-election analysis shows that Kerry won beyond a reasonable doubt. The analysis is based on plausible and feasible estimates of 1) total votes cast in 2000 and 2004, 2) 2000 voter mortality, 3) turnout of living Bush and Gore voters in 2004, and 4) 12:22am National Exit Poll vote shares. The analysis indicates that Kerry won by 53-46% (10 million)- a 13 million margin discrepancy from the recorded vote.

Why use 12:22am (13047 respondents) and not the Final NEP (13660) vote shares? Because the Final NEP vote shares were forced to match the vote and were not plausible. Just changing to a mathematically impossible Bush/Gore returning voter mix was insufficient to match the recorded vote; the vote shares had to be changed as well. It was the equivalent of the Texas two-step: the Final was forced to match the recorded vote by 1) changing the returning voter mix (even though it required 6 million phantom Bush voters) and 2) increasing Bush’s corresponding 2004 vote shares beyond the margin of error.

False Recall was the ultimate naysayer “explanation” for the Final 2004 National Exit Poll anomalies. The 2006 and 2008 Final NEP anomalies exposed the fallacy of the 2004 argument. A True Vote analysis based on the recurring 2004, 2006 and 2008 anomalies finally puts “false recall” to eternal rest.

1) Bush approval was in the tank: 35% (2006) and 22% (2008). It was 48% in 2004.
2) The returning Bush/Kerry voter mix was changed from an improbable ratio in the unadjusted exit poll to an impossible one in the Final NEP.
3) Millions of phantom Bush voters were required to match the recorded vote.

It should be obvious to any impartial observer that there is no basis for comparison between the limited NES samples and the massive pre-election and exit poll samples. The evidence is overwhelming that Kerry won the election. The only question is by how much.

The final 18 national pre-election polls (after undecided voters were allocated) projected that Kerry would win by 51-48%.
The unadjusted state exit poll aggregate (WPE/IMS) indicated he won by 52-47%.
The initial adjusted aggregate (“Best GEO”) had 51-48%.
The final (12:40am) adjusted aggregate (“Composite”) had 50.5-48.5%.

The True Vote Model indicates that Kerry won by 53-46%, a 10 million margin. The Final National Exit Poll (Bush by 50.7-48.3%)was forced to match a fraudulent recorded vote using an impossible number of returning Bush 2000 voters.

In the 2006 midterms, the Final NEP indicated there were 5 million more returning Bush voters than Kerry voters. Bush “won” the in 2004 by 3.0 million recorded votes. The preliminary 730pm NEP had a 47/45 returning Bush/Kerry voter mix and projected a 55-43% Democratic landslide. The Final indicated a 49/43 mix and a 52-46% Democratic margin. Of course, the Final was forced to match the recorded vote.

The Final 2008 NEP shows a ridiculous 46/37 returning Bush/Kerry mix. Preliminary state and national exit polls have not been and apparently never will be released. But we don’t need them. As in 2004, the returning voter mix was mathematically impossible; it indicated 12 million more returning Bush than Kerry voters.
Even if one believes that the 2004 election was fraud-free, how could there have been 12 million more returning Bush voters? On the other hand, if Kerry won by the unadjusted exit poll 52-47% split (a six million vote margin) one would expect approximately 6 million more returning Kerry voters than Bush voters - a whopping 18 million net deviation.

To explain the returning voter mix, will Lindeman claim that returning Kerry voters misreported their 2004 vote and say they voted for Bush - even though he had a 22% approval rating on Election Day 2008? That would be the only way to explain the official 9.5 million Obama margin (52.9-45.6%). Apparently, the 2004 vote-switching algorithm could not overcome a 22 million Obama True Vote landslide.

The media consortium that sponsored the exit polls has not released the 2008 exit poll report. The 2004 report was released on Jan.19, 2005, one day before the Bush inaugural. It is clear that the consortium does not want a repeat of the 2004 exit poll controversy. Releasing the state unadjusted exit poll (WPE) data would only confirm that election fraud is endemic.

The “false recall” hypothesis used to explain the impossible Final 2004 National Exit Poll 43/37% returning Bush/Gore voter mix is just, well, false.

1. False recall assumes that Alzheimer's was a Gore voter malady. Bush voters were immune to "mere forgetfulness" or a “slow drifting fog”.

2. Comparing the difference between the recorded vote and the retrospective past vote is a false dichotomy. The recorded vote, by definition, cannot represent the True Vote since it does not include millions of uncounted votes (spoiled, absentee, late and provisional ballots) that are heavily (70-80%) Democratic. Only by measuring the past vote against the True Vote (i.e. Total Votes Cast) is the analysis even remotely valid.

3. The average NES/True vote share deviation was 0.40%. Where’s the beef?

4. Comparing the True Vote average margin to the NES average survey margin for the 11 elections from 1968-2008 debunks "false recall" and confirms endemic election fraud. Apparently, Lindeman never considered that the NES study would confirm the True Vote Model because he never did the analysis. NES results must be compared to the True Vote (i.e. total Votes Cast) for the analysis to have any validity.

5. False recall is based on the premise that the returning Bush/Gore voter split (43/37%) in the Final National Exit Poll was a sampled result and that 3 of 40 (7.5%) Gore voters misreported their past vote. But it's standard operating procedure of the exit pollsters to force the Final to match the recorded vote by adjusting selected weights and/or vote shares in all demographic categories.

6. Even if we assume that the Final NEP (13660 respondents) returning Bush/Gore voter split was a true sample, how does that explain that at 12:22am (13047 respondents) the split was 41/39%? Surely the 4% change in weights for 613 additional respondents is implausible. And only 3182 respondents were even asked the question about their past vote.

7. The majority (10,478) of respondents were only asked whom they voted for minutes before - not 4 years ago. And most said Kerry. Surely five minutes was too short a period for "mere forgetfulness", "slow-moving fog' or "bandwagon effect" (Bush had a 48% approval rating). The margin of error for a 10,000 sample is less than 1%.

8. In addition to the Final NEP, there were 50 state exit polls. The unadjusted aggregate national share was 52-47% in favor of Kerry. False recall was proposed to explain the Final NEP past vote anomaly, but ignores the state exit polls in which the respondents were asked who they just voted for. The margin of error for the total 70,000 state sample is significantly lower than 1% - even after adding an incremental "cluster effect".

9. The 2006 midterm and 2008 Final NEP effectively buried "false recall". Of course, it's standard operating procedure that the returning Bush/Kerry voter mix (as well as the vote shares) would be adjusted to match the recorded vote. But look at how they were adjusted. In the 2006 midterms, the Bush/Kerry mix was changed from 46.5/45.5% in the unadjusted exit poll to 49/43% in the Final, cutting the Democratic margin from 56-42% to 52-46%. Bush had a 35% approval rating in 2006. The 2008 NEP returning Bush/Kerry mix (46/37%) was mathematically impossible; it implied that there were 12 million more returning Bush than Kerry voters. Bush "won" in 2004 by a fraudulent 3.0 million margin. He won the recorded vote by 52.9-45.6%. We don’t have the unadjusted exit polls (will they ever be released?), but a True Vote analysis indicates that Obama won by 57-41%.

10. To say that a "slow-drifting fog" was unique to Gore voters is laughable and should be dismissed out-of-hand. Lindeman was left with a "latent Bush bandwagon effect" and an "incumbency advantage" as possible explanations. But the 12:22am NEP indicated that 10% of returning Bush voters defected to Kerry and only 8% of Gore voters defected to Bush.

11. It strains credulity that Gore voters would vote for Kerry and lie about their past vote. Or would forget that the election was stolen from them. How could they forget that the Supreme Court nullified the Florida Supreme Court, halted the recount and handed the election to Bush?

12. But the bottom line is that it's irrelevant whether or not Gore voters misreported. Regardless of how they said they voted, the only thing that matters is a) how they actually voted in 2000 and b) how they actually voted in 2004. Gore had 51.0 million recorded votes and Bush 50.5 million. For simplicity, we can ignore the 5.4 million net uncounted votes (70-80% were for Gore). The recorded 2000 vote and the 1222am NEP indicated that Gore and Kerry were winners.

13. The Final NEP 43/37 weightings did not consider that even Bush 2000 voters were mortal (5% died prior to 2004). Besides, the fact is that voter turnout is always less than 100%. The “Clincher” sensitivity analysis shows that even assuming the unlikely scenario of 98/90% returning Bush/Gote voter turnout, Kerry still wins by 7 million votes. With equal turnout, Kerry wins by 10 million.

14. The Aug. 2005 "Game" was a classic Democratic Underground online debate between spreadsheet-wielding Internet bloggers and two prominent exit poll naysayers (Lindeman and Elizabeth Liddle). Prior to the "Game", Lindeman was challenged time and again to provide a plausible scenario to show how Bush "won" the election by 3.0 million. He agreed to the stipulation that we are all mortal and reduced the number of Gore and Bush voters by the annual mortality rate. To support a non-biased analysis for the base case scenario, equal Bush and Gore voter turnout rates were assumed. Lindeman effectively abandoned "false recall" in agreeing to a feasible, plausible returning voter mix.

15. But Lindeman could not match the recorded 2004 vote using the Final NEP Bush vote shares that had already been inflated to match the recorded vote. His only recourse was to inflate the shares even further - far beyond the margin of error. The "Game" proved that he could not provide one plausible Bush win scenario. But Lindeman did not quit; he has since reverted back to “false recall” which he now uses to explain all 2004, 2006 and 2008 Final NEP returning Bush voter anomalies. And now you’ve heard the rest of the story.

Postscript
On page 4 of his paper, Lindeman writes:
“Consider: In the 2000 VNS exit poll, the weighted results indicate that 45.6% of respondents had voted for Bill Clinton in 1996, while 31.3% had voted for Bob Dole, 6.5% for Ross Perot, 2.4% for another candidate, and 12.5% had not voted (1.7% skipped the question). The unweighted results are similar. 9 Thus, Clinton’s apparent margin over Dole is 14.3% of all respondents, or 16.7% of those who said they had voted in 1996. By contrast, in the 1996 returns, Clinton received 49.2%, Dole 40.7%, Perot 8.4%, and other candidates about 1.7%. That is, Clinton won by only 8.5 points. Of course, we cannot assume that the 2000 proportions should match the 1996 proportions. Nonetheless, these results are “impossible” in precisely the same sense as the 2004 results. In 2000, 105.4 million presidential votes were counted, of whom 45.6% would equal about 48.1 million Clinton 1996 voters. Yet Clinton received only 47.4 million votes in 1996, of whom presumably under 46 million survived in 2000. The maximum possible proportion of Clinton 1996 voters in the 2000 electorate is some two points lower than the reported proportion – a discrepancy well beyond expected sampling error.10

How, then, do we explain the “impossibly high” proportion of Clinton 1996 voters in the 2000 exit poll? Is it likely that Al Gore stole millions of votes in 2000, and Clinton voters were upweighted in the exit poll to match the result? Or that millions of votes were stolen from Clinton in 1996? Perhaps anything is possible, but given that the 1996 and 2000 national exit poll estimates came close to the official returns, neither conjecture seems at all likely. Likelier, again, is that many respondents in 2000 wrongly reported having voted for Clinton in 1996”.

That’s a reasonable question. But once again, Lindeman makes the “mistake” of using the recorded vote- not Total Votes Cast. He ignores uncounted votes. The 2000 Final National Exit Poll as always, was forced to match the recorded vote. The True Vote is based on the unmentionable Total Votes Cast.

In 2000, there were 110.4 million votes cast and 105.4 million recorded.
There were 5.4 million net uncounted votes.
In 1996, there were 105.2 million votes cast and 96.3 million recorded.
There were 8.9 million net uncounted votes.

Let’s include uncounted votes in order to determine the True Vote in both elections. The best estimate is that 75% of the uncounted votes were for Clinton. Therefore, there were 54.0 million votes cast for Clinton, 41.2m for Dole, 9.9m for Perot/other.

Deduct 5% from votes cast for mortality. Next assume that 96% of voters turned out in 2000. There were 51.4m returning Clinton voters, 35.6m Dole and 8.8m Perot/other and 15.9m DNV96. The returning voter mix was 48.7% Clinton, 33.8% Dole and 8.3% Perot/other. The Clinton 14.9% voter turnout margin is within 0.6% of the Final National Exit Poll (14.3%) turnout margin!

[code]

1968-2008 Comparison of Recorded and True Vote Margins to NES Retrospective Surveys

The True Vote shares were within 1% of NES!

NES vs. True Vote Share (1968-2008)
The average absolute deviation for the 11 elections was -0.40%.
The average Democratic absolute deviation was -0.70%.
The average Republican deviation was 0.46%.

NES vs. Recorded Vote (1968-2008)
The average absolute vote share deviation was -1.75%.
The average Democratic absolute deviation was -3.30%.
The average Republican deviation was -0.46%.

Note:
2008 stats based on the average of the following retrospective margins:
WSJ 12.0; NYT 30.0


1968 1972 1976 1980 1984 1988 1992 1996 2000 2004 2008 Average
Winner Nixon Nixon Carter Reagan Reagan Bush Clinton Clinton Bush Bush Obama
R R D R R R D D R R D

NES

Total -6.7% -28.3% 2.3% -11.4% -16.3% -5.7% 13.9% 15.5% 4.2% -0.4% 21.0% -1.08% -0.54%
ABS 6.7% 28.3% 2.3% 11.4% 16.3% 5.7% 13.9% 15.5% 4.2% 0.4% 21.0% 11.43% 5.71%
Dem 2.3% 13.9% 15.5% 4.2% 21.0% 11.38% 5.69%
Repub 6.7% 28.3% 11.4% 16.3% 5.7% 0.4% 11.47% 5.73%

True Vote

Total 3.4% -16.9% 6.8% -6.9% -13.4% 2.1% 21.0% 16.5% 4.3% 8.1% 17.6% 3.88% 1.94% 2.48%
ABS 3.4% 16.9% 6.8% 6.9% 13.4% 2.1% 21.0% 16.5% 4.3% 8.1% 17.6% 10.63% 5.32% -0.40%
Dem 3.4% 6.8% 2.1% 21.0% 16.5% 4.3% 8.1% 17.6% 9.98% 4.99% -0.70%
Repub 16.9% 6.9% 13.4% 12.39% 6.19% 0.46%


Recorded Vote

Total -0.7% -24.4% 2.1% -9.7% -18.2% -7.7% 5.6% 8.5% 0.5% -2.5% 7.3% -3.58% -1.79% -1.25%
ABS 0.7% 24.4% 2.1% 9.7% 18.2% 7.7% 5.6% 8.5% 0.5% 2.5% 7.3% 7.93% 3.96% -1.75%
Dem 2.1% 5.6% 8.5% 0.5% 7.3% 4.79% 2.39% -3.30%
Repub 0.7% 24.4% 9.7% 18.2% 7.7% 2.5% 10.54% 5.27% -0.46%

4% MoE
Rec-NES 3.0 2.0 -0.1 0.9 -0.9 -1.0 -4.2 -3.5 -1.9 -1.1 -6.9 -1.25
9< MoE Yes Yes Yes Yes Yes Yes No Yes Yes Yes No Yes

True-NES 5.0 5.7 2.3 2.3 1.5 3.9 3.6 0.5 0.0 4.3 -1.7 2.47
8 < MoE No No Yes Yes Yes Yes Yes Yes Yes No Yes Yes

Rec-True 2.0 3.8 2.4 1.4 2.4 4.9 7.7 4.0 1.9 5.3 5.2 3.72
6 <MoE Yes Yes Yes Yes Yes No No No Yes No No Yes

Final NEP Returning Voter Mix (past vote forced to match)
1968 1972 1976 1980 1984 1988 1992 1996 2000 2004 2008 Average
Dem 55% 35% 30% 42% 34% 33% 28% 38% 41% 37% 37% 37.3%
Rep 35% 44% 53% 41% 44% 58% 53% 31% 33% 43% 46% 43.7%
Margin 20% -9% -23% 1% -10% -25% -25% 7% 8% -6% -9% -6.5%

____________________________________________________________________

Final 2000 National Exit Poll
(forced to match the recorded vote)

95% 96%
Cast Record Alive Voted Voter
1996 1996 1996 2000 2000 Turnout Mix Gore Bush Other Turnout/Alive
DNV - - - - 15.92 15.1% 42% 51% 7% -
Clinton 49.2% 53.95 47.40 45.03 43.23 48.07 45.6% 82% 15% 3% 107%
Dole 40.7% 41.16 39.20 37.24 35.75 33.00 31.3% 7% 91% 2% 89%
Perot 10.1% 9.90 9.68 9.20 8.83 8.43 8.0% 31% 61% 8% 92%

Total 105.02 96.28 91.47 87.81 105.42 Total 48.41% 47.90% 3.69%
51.03 50.50 3.89

Recorded 48.38% 47.86% 3.76%
105.42 51.00 50.46 3.96


True Vote Model
95% 96%
Cast Record Alive Voted Voter
1996 1996 1996 2000 2000 Turnout Mix Gore Bush Other Turnout/Alive
DNV - - - - 15.05 9.1% 42% 51% 7% -
Clinton 53.7% 56.34 47.40 53.53 51.39 51.39 48.7% 82% 15% 3% 96%
Dole 37.2% 39.03 39.20 37.08 35.60 35.60 33.8% 7% 91% 2% 96%
Perot 9.2% 9.64 9.68 9.16 8.79 8.79 8.3% 31% 61% 8% 96%

Total 105.02 96.28 99.77 95.78 110.83 Total 48.76% 47.79% 3.76%
54.04 52.97 3.82

Recorded 48.38% 47.86% 3.76%
105.42 51.00 50.46 3.96

____________________________________________________________________

2004

12:22am National Exit Poll
13047 respondents

105% turnout of Bush 2000 voters
Kerry wins by 4.5 million votes

Voted Mix Kerry Bush Other Turnout/Alive
DNV 17% 57% 41% 2% -
Gore 39% 91% 8% 1% 98.4%
Bush 41% 10% 90% 0% 104.6% ?
Other 3% 64% 17% 19% 97.5%

NEP 100% 51.20% 47.50% 1.30% 3.70%
Votes 122.30 62.62 58.09 1.59 4.52

Recorded 122.30 48.27% 50.73% 1.00% -2.46%
Votes 122.30 59.03 62.04 1.23 -3.01


Final National Exit Poll
Posted 1:25pm on CNN
13660 respondents
(forced to match the the recorded vote)

Requires 110% turnout of Bush 2000 voters!

Mix Kerry Bush Other Turnout/Alive
DNV 17% 54% 45% 1% -
Gore 37% 90% 10% 0% 93.4%
Bush 43% 9% 91% 0% 109.7%
Other 3% 71% 21% 8% 97.5%

NEP 100% 48.48% 51.11% 0.41% -2.63%
Votes 122.30 59.29 62.51 0.50 -3.22

Recorded 100% 48.27% 50.73% 1.00% -2.46%
Votes 122.30 59.03 62.04 1.23 -3.01


True Vote Analysis

Base Case Assumptions:
Gore won 75% of 5.4 million net uncounted votes (majority in minority districts)
Voter Mortality: 5% over the four year period 2000-2004 (accurate)
Voter Turnout: 98% of LIVING Gore and Bush voters (Census/new voter turnout)

Vote Shares: based on 12:22am NEP
Note: Kerry had 62% of DNV at 4pm; 59% at 730pm, 57% at 1222am, 54% in the Final.
Final NEP shares were not used as they were FORCED TO MATCH THE RECORDED VOTE.
Kerry had 62% of DNV at 4pm; 59% at 730pm, 57% at 1222am, 54% in the Final (see NEP).

Kerry won the True Vote by 10.2 million (53.4-45.3%).

95% 98%
Cast Record Alive Voted 2004 True Vote
2000 2000 2000 2004 2004 Turnout Mix Kerry Bush Other Turnout/Alive
DNV - - - - 22.56 15.6% 57% 41% 2% -
Gore 55.79 51.00 53.00 51.94 51.94 42.5% 91% 8% 1% 98.0%
Bush 51.05 50.46 48.50 47.53 47.53 38.9% 10% 90% 0% 98.0%
Other 3.99 3.96 3.79 3.71 3.71 3.0% 64% 17% 19% 98.0%

Total 110.83 105.42 105.28 103.18 125.74 100% 53.39% 45.30% 1.31% 8.09%
Cast Recorded Total 67.13 56.96 1.65 10.17
Gore 50.34% 48.38% Recorded 48.27% 50.73% 1.00%
Bush 46.06% 47.86% 122.30 59.03 62.04 1.23
Other 3.60% 3.76%



Sensitivity Analysis 1

2000 Voter Turnout in 2004

Kerry wins by 7.3 million in the worst case scenario.
(98% Bush voter turnout; 90% Gore voter turnout).

He wins by 10.2 million assuming equal Gore/Bush 98% turnout.

Kerry Share
Bush Gore Turnout
Turnout 90% 92% 94% 96% 98%

90% 53.7% 54.0% 54.3% 54.6% 54.9%
92% 53.3% 53.6% 53.9% 54.2% 54.5%
94% 53.0% 53.3% 53.5% 53.8% 54.1%
96% 52.6% 52.9% 53.2% 53.5% 53.8%
98% 52.2% 52.5% 52.8% 53.1% 53.4%

Kerry Margin

90% 11.1 11.8 12.5 13.3 14.0
92% 10.1 10.9 11.6 12.3 13.0
94% 9.2 9.9 10.6 11.4 12.1
96% 8.2 8.9 9.7 10.4 11.1
98% 7.3 8.0 8.7 9.4 10.2



Sensitivity Analysis 2

Voter Mortality and 2000 Voter Turnout in 2004
Bush voter turnout assumed equal to Gore turnout

Key result:
Mortality rate has virtually ZERO impact on vote share.
(1% change in mortality results in .04% change in vote share)

Kerry Share

Mortality 2000 Voter Turnout
(4-year) 96% 97% 98% 99% 100%

6.0% 53.14% 53.28% 53.43% 53.57% 53.72%
5.5% 53.11% 53.26% 53.41% 53.55% 53.70%
5.0% 53.09% 53.24% 53.39% 53.54% 53.68%
4.5% 53.07% 53.22% 53.37% 53.52% 53.67%
4.0% 53.05% 53.20% 53.35% 53.50% 53.65%

Kerry Margin

6.0% 9.6 9.9 10.3 10.6 11.0
5.5% 9.5 9.9 10.2 10.6 11.0
5.0% 9.4 9.8 10.2 10.5 10.9
4.5% 9.4 9.8 10.1 10.5 10.9
4.0% 9.3 9.7 10.1 10.4 10.8



2004 National Exit Poll Timeline

AGE Demographic
(Weights and vote shares in percent)
<< indicates percent change from 12:22am

Note that there is very little change from 4pm to 12:22am.
But the 30-44 and 60+ category weights changed by 2% in the Final.
Bush was leading in the 2 groups at 12:22am! Just a coincidence?
Kerry vote shares were reduced by 2-3% in all age groups in the Final!

There were only 613 additional respondents after 12:22am.

ALL OTHER DEMOGRAPHIC CATEGORIES HAD TO BE ADJUSTED IN THE FINAL
NATIONAL EXIT POLL TO FORCE A MATCH TO THE RECORDED VOTE.


Update 4pm 730pm 1222am Final 4pm 730pm 1222am Final
Respondents 8349 11027 13047 13660 8349 11027 13047 13660

AGE
Category Weights (%) Kerry vote share (%)
18-29 15 17 17 17 56 56 56 54 <<
30-44 27 27 27 29<< 48 49 49 46 <<<
45-59 31 30 30 30 52 51 51 48 <<<
60+ 27 26 26 24<< 48 48 48 46 <<

Total 100 100 100 100 50.44 50.53 50.53 47.96

VOTE (in millions) 61.67 61.78 61.78 58.64