2004 general election

Jump to: navigation, search

The 2004 presidential election was fought between George W. Bush and John Kerry. On election night, the exit polls gave Kerry a decisive win, but the election results turned out in Bush's favor, giving him the presidency. This massive exit poll discrepancy was shocking and unprecedented in US politics. Exit pollsters were quick to find errors in their own work, but election integrity analysts came to the opposite conclusion: exit polls were right and vote counts were wrong. Over the next several years, investigators would find election irregularities in many states, most notably Ohio.

Exit polls

Exit polling by Edison/Mitofsky (E/M) showed Kerry winning the popular vote by 3.0%, while the official results had Bush winning by 2.5%. This 5.5% red shift, which saw the exit polls flipped on their head, was an unprecedented exit polling discrepancy in recent US history. A similar red shift occurred in nearly every state, and was curiously more pronounced in swing states. Ohio and New Mexico both flipped from Kerry to Bush, and Florida went from a virtual tie to a solid Bush win.

The magnitude and persistence of the red shift made this impossible to attribute to random error. Either the exit polls were biased towards Kerry, the official results were inaccurate, or some combination thereof.

2004's exit poll discrepancy continued a trend that began around 2000. The 2000 Florida election had a 6.6% red shift, and was marred by election irregularities mainly affecting Gore, including tampered electronic vote counts. Multiple states in 2002 bore signs of electronic vote rigging, and the exit polls were so far off that the media withheld them. The development of a pervasive red shift coincided with the proliferation of electronic voting machines that made wholesale election fraud possible. That, combined with the heightened red shifts in swing states (more likely targets of rigging), made people suspect the 2004 red shift indicated fraud.

E/M, however, rejected the possibility of election fraud. They released a report claiming that differential nonresponse was the cause of the discrepancies. Analyzing the exit polls, E/M found that representative precincts were sampled, but there was a disparity between voter responses and official results in those precincts ("within-precinct error", or WPE). They explained the WPE on the basis of Kerry voters being more likely than Bush voters to complete exit polls. E/M asserted that response rates of 56% for Kerry voters and 50% for Bush voters could account for all the discrepancies. This came to be called the "reluctant Bush responder" (rBr) hypothesis. E/M didn't release the exit poll surveys to support their conclusion, but they did release response rates broken down by precinct partisanship.[1]

US Count Votes (USCV), a group of election integrity analysts, disputed E/M's conclusion. They released their own report on March 31, 2005, which found[2]:

  • Individual states' exit poll discrepancies formed a normal distribution, as expected, but with the mean shifted from 0% to pro-Kerry. This means that there was either a consistent pro-Kerry polling bias or vote miscount.
  • Response rates in high-Bush precincts (80% of the vote or greater) were slightly higher than those in high-Kerry precincts. This is the opposite of what rBr should produce.
  • One might rebut the above by saying that Bush response rate was high in high-Bush precincts (due to the preponderance of other Bush voters), but low in the presence of Kerry voters. If that were true, exit polls should be most accurate in high-Bush precincts, since response bias was lower and exit poll participation was higher. But in reality, the exit polls became less accurate as the share of Bush voters increased.
  • To match E/M's response rates and WPE in the five precinct partisanship groups, implausible Kerry and Bush response rates were needed. Bush voters would have needed to have slightly higher response rates in high-Kerry precincts, and Kerry voters would have needed to have massively higher response rates across all others, especially the high-Bush precincts. This runs completely contrary to common sense.
  • For high-Bush precincts, the mean WPE was -10.0% and the median WPE was -5.9%. This means that in half of the high-Bush precincts, the WPE was at least -14.2%. Necessary Kerry and Bush response rates for those precincts are either highly implausible (Kerry has about a 29% higher response rate) or completely impossible (Kerry's response rate is over 100%).

In short, the 2004 exit poll data was highly inconsistent with E/M's rBr hypothesis.

Elizabeth Liddle, a collaborator with USCV at the time, also noted that a constant response bias across precincts of varying partisanship would produce an "inverted U" WPE pattern. WPE would be at a minimum in 100% Kerry or Bush precincts, and at a maximum in evenly-split precincts. This made it even less likely that rBr explained the exit poll discrepancy, since WPE was highest in the high-Bush precincts, not at a minimum.

Liddle, however, then wrote a paper asserting that the E/M data was consistent with rBr. She simulated precincts of varying partisanship, with response rates normally-distributed around a 56% Kerry/50% Bush mean, and claimed to replicate the E/M data. Another one of her contributions was a formula to convert WPE into a response bias measure. This would, for instance, turn the "inverted U" into a flat line.

Aside from the absurdity of Liddle claiming the E/M data fit with rBr when there was no "inverted U", USCV found that Liddle's analysis did not completely replicate the E/M data, and required dropping 10% of the high-Bush precincts.[1] They ran some of their own WPE simulations to check Liddle's work, calculating probabilities, and found rBr to be a nearly impossible explanation for the E/M data. USCV also calculated the effect of vote shifting along with response bias that varied logically by precinct partisanship (Kerry bias in precincts Kerry won, and Bush bias in precincts Bush won), and was able to fit the E/M data fairly well.[3]

Nevertheless, Warren Mitofsky of E/M embraced Liddle's analysis and hired her to help him prove the differential nonresponse theory. At the American Association of Public Opinion Research (AAPOR) conference in 2005, Mitofsky displayed two scatterplots of the exit poll data. He first showed the WPEs by precinct partisanship, which went in both pro-Kerry and pro-Bush directions (likely due to various survey error factors) but trended towards pro-Kerry. Mitofsky then used Liddle's bias index in place of WPE, showing no correlation with partisanship. The implication was that there was a constant response bias, supporting rBr.[1]

This made little sense, as a constant response bias had already been disproven by USCV from the E/M aggregate data. A zero correlation, in any event, proved nothing, since it could have also been generated by nonlinear variations. In fact, Mitofsky soon retracted the rBr theory of a constant response bias in private correspondence with USCV member Ron Baiman. He implicitly agreed that a constant bias couldn't explain the data, and it had to vary by precinct partisanship, but argued the overall bias was still 56% Kerry to 50% Bush. E/M refused, however, to release a multivariate regression proving this.

USCV followed up with a September 8 report noting that this theory still failed to explain the discrepancies. They referred back to calculations from their March 31 report, showing that required overall Kerry and Bush response rates in precinct quintiles were implausible, particularly in high-Bush precincts where the required bias was 84% Kerry/52.9% Bush. Even trying to minimize the response bias as much as possible, this implausibility remained, and the required overall average for the entire exit poll would have been 58%/50% rather than 56%/50%. Furthermore, they explained why the E/M data was consistent with vote shifting as per their simulation: flipped votes would cause precincts with large WPEs to shift into higher Bush precincts, explaining the larger WPE as Bush vote share increased.[4]

Aside from this academic back-and-forth, USCV had noted other evidence that the exit polls indicated vote miscount:

  • In the exit poll crosstabs, a plurality of voters said they had voted for Bush in 2000 even though Gore won. This indicated that if anything, the polls oversampled Republicans.[1] A "false recall" theory, saying that voters misremembered their 2000 vote and said they chose the incumbent, might partially explain it, but it's weakened by Bush's sub-50% approval and the higher profile of 2000 compared to other elections.
  • E/M's data showed that the WPE was substantial across all precincts except those using hand-counted paper ballots, the gold standard of election accuracy. High discrepancies when using corruptible voting technology, and low discrepancies when using a verifiable counting process, is very telling.[2]
  • The 2004 exit polls had a large red shift for the presidential race, but were much more accurate for the Senate races. If Republicans were underrepresented, as the differential nonresponse theories stated, both races should have been similarly affected.[2]

Following this debate, differential nonresponse causing the 2004 exit poll discrepancy was soundly debunked. Still, Mitofsky and others advanced another argument against election fraud: if precincts exhibited election fraud for Bush in 2004, he should have done better in those precincts than in 2000. USCV identified numerous flaws in this "swing vs. red shift" argument, and mathematically proved it to be of no analytical value[1]:

  • There are infinite ways for a candidate to win fraudulently without improving their vote share from 2000.
  • 2004 vote share in a precinct is not necessarily tied to the 2000 vote share. Political factors can easily change things, and the precinct boundaries themselves might have even changed between 2000 and 2004.
  • Comparing vote share says nothing about the weights of the precincts.
  • Doing this entire test assumes the 2000 results were correct. They were clearly not all correct, especially in Florida.

To truly settle the matter, USCV had repeatedly implored E/M to release the raw exit poll data, to no avail. Mitofsky claimed that doing so would have compromised respondent confidentiality, under the unlikely possibility that anonymous questionnaires would be traced back to voters. Yet he gave the raw data to supporters of his exit poll error theory, including Liddle and the Election Sciences Institute (ESI), which purported to debunk fraud in the 2004 Ohio election with the above argument. Using the raw data from the ESI report, USCV released an analysis definitively proving that the 2004 Ohio exit poll discrepancies were due to vote miscount. This direct proof of vote miscount based on a single state's raw data only strengthened the argument that the same was true nationwide.[1]

General issues

Voter suppression

Retail vote rigging

Electronic vote rigging



Main article: 2004 Ohio general election

Just like Florida in 2000, Ohio was the focal swing state in the 2004 presidential election. GOP election officials and voting system contractors engaged in many forms of election manipulation, ranging from dirty tricks to outright fraud, to ensure Bush won the state over Kerry. Thousands of eligible Democratic voters were disenfranchised due to voter registration irregularities and voter suppression. Voting systems, meanwhile, were manipulated to reverse the election outcome to Bush. Private companies such as ES&S and Triad tampered with central tabulators, and the vote counts were routed through partisan out-of-state servers that left them vulnerable to remote alteration. A recount was similarly corrupt, with ballots altered to match the electronic count, and election officials and vendors colluding to prevent discrepancies with the machines from being discovered.


There was evidence of electronic vote rigging for Bush in Florida on both touchscreens and optical scanners.

A study by UC Berkeley students found that touchscreens added up to 260,000 votes to Bush's margin, with the effect mainly concentrated in heavily-Democratic South Florida: Broward, Palm Beach, and Miami-Dade counties.[5][6] The EIRS incident database confirms that most reported voting machine issues in Florida occurred in those same three counties.[7] Yang Enterprises (YEI) whistleblower Clint Curtis later claimed that Tom Feeney was interested in manipulating touchscreen voting machines to "control the vote in South Florida". No independent proof of this exists, but Curtis's other allegations against YEI and Feeney are documented to be true.

Optical scanners also showed signs of fraud. When Bev Harris went to acquire Volusia County's poll tapes, she was obstructed by election officials, was given newly-printed copies rather than the originals, and caught them trying to throw away the originals.[8] A look at the poll tapes showed anomalies indicating that the memory cards, which held the code to tabulate the votes, had been tampered with.[9] Incident reports from EIRS additionally found that optical scanners in Leon and Brevard counties (the same type as in Volusia) failed on election day[7], which could again indicate memory card tampering.

New Mexico

New Mexico had an unusually high percentage of undervotes, which were tied to the use of DRE machines and concentrated in minority areas. 44% of the optical scan precincts had 0 undervotes, while only 2% of the pushbutton DRE precincts did. 89 precincts had a more than 10% undervote rate, with all but one of them using pusbutton DREs. These differences were found by USCV to be statistically-significant. USCV's analysis also showed that the DRE problems mainly impacted regions with higher populations of ethnic minorities. Within those 89 high-undervote precincts, the minority population was 80%, compared to 50% in the rest of the state. And the best predictor of undervote rate was a combination of machine type and minority population. This implies that pushbutton DREs were specifically rigged in minority areas to produce more undervotes.[10]

North Carolina






  1. 1.0 1.1 1.2 1.3 1.4 1.5 National Election Data Archive, "History of the Debate Surrounding the 2004 Presidential Election", 2005/11/05 - 2004 exit poll analysis overview by the US Count Votes team
  2. 2.0 2.1 2.2 US Count Votes, "Analysis of the 2004 Presidential Election Exit Poll Discrepancies", 2005/04/12 - debunking "reluctant Bush responder" (rBr)
  3. US Count Votes, "Patterns of Exit Poll Discrepancies", 2005/05/21 - debunking Mitofsky and Liddle's response with simulations
  4. US Count Votes, "The 2004 Presidential Election: Exit Poll Error or Vote Miscount?", 2005/09/08 - debunking overall average response bias
  5. Oakland Tribune, "Study Finds Florida 'Ghost' E-Votes", 2004/11/19
  6. "Working Paper: The Effect of Electronic Voting Machines on Change in Support for Bush in the 2004 Florida Elections" by Michael Hout, Laura Mangels, Jennifer Carlson, and Rachel Best - analysis of Florida touchscreen vote irregularities
  7. 7.0 7.1 EIRS incident reports for 2004 Florida
  8. Bev Harris, "Vote Fraud, Volusia County On Lockdown", 2004/11/18 (DU thread) - Volusia County FL poll tapes PRR dispute
  9. Black Box Voting, "11/04 - Volusia: Poll Tapes Voting Machine Results Reports", 2005/12/22 - signs of memory card tampering in Volusia County
  10. US Count Votes, "Analysis of Undervotes in New Mexico’s 2004 Presidential Ballots", 2005/01/03 - New Mexico undervotes correlated with DRE use
  11. North Carolina absentee discrepancies
  12. Charlotte Observer, "Officials visit Monday to review counting and auditing procedures", 2004/11/19
  13. Charlotte Observer, "Gaston elections director takes leave", 2004/11/24

External links

Exit poll study

Voting irregularities

  • EIRS incident reports
  • Volusia County FL tampering
    • BBV forum post in December 2004 about ActionSpeaksPortland
      • "You hear about situations that were confrontational, but out of hundreds of visits to elections officials, how many involved confrontation?

        Theresa LePore, to be sure. But then, she put crime tape up around the elections office to prevent anyone from entering on Nov. 2, and she counted ballots two weeks BEFORE the election. She point-blank refused public records requests. Her behavior justified confrontation by the citizenry.

        Volusia County. But then, they were throwing away election records, tampering with memory cards, making homemade ballots, taking voter documents to their personal homes. Their behavior warranted confrontation by citizens."
      • "The biggest problem, though, is still in the counting. You still use electronics to count and tally your votes. Whether those electronic counts are secure is still anyone's guess. The Diebold mail-in vote counting machines have been stripped of a key safeguard, the poll tape, relying entirely on the defective GEMS system to keep a record of votes."
      • "I like unions. But making it hard to fire election officials can be problematic -- in Texas, at one point, they had so much trouble with Conny [maiden name Drake] McCormack and her alleged election-rigging that they had to introduce a bill in the legislature to make it easier to fire election officials. The bill was known as the "Get Conny Drake" bill. This was around 1985, but the issue would be similar today."
      • "Er...what about secure chain of custody of who gets a paper ballot in the first place, how many paper ballots are produced (witness the Diebold 25% "spoilage" [read: overprinting] rate)."
    • Black Box Voting, "Volusia County Update", 2005/02/21
    • Black Box Voting, "11/04 - VIDEO: Garbage raid at the Volusia County warehouse", 2005/06/23
      • "By the way, while it's true there were sample ballots in the bag, not all were samples. What we found was a whole bunch of half-sized ballot masters. Put on the machine at 200% and copied they turn into ballots (except the ballot is missing a serial number). So, I got to wondering: Does this trash contain any un-serialnumbered full sized ballots that do not say "sample" that match Fidlar Doubleday ballot masters found in the same bag (curiously, Fidlar uses mini-ballot masters for packing slips, how handy!)

        The answer was yes. They were making ballots in the warehouse.

        Add to this: The Volusia County ballot vault was available to the some same people who were at the warehouse, 24 hours a day. Put these two things together and the hand-count audits become a joke."
      • "anwar: some of the same people. Specifically, we have very credible reports from a Volusia County official that Lana Hires had full access to both the ballot vault and the warehouse, and it is not unusual for the ballot vault there to be left unlocked with the door standing open. In fact, we observed that ourselves, meaning anyone at the elections headquarters might be able to nip in there for a quick ballot swap, especially if you have some minty fresh ones printed at the warehouse.

        We observed "Pete" from the warehouse also at the elections headquarters where the ballot vault is, and we also saw Denise Hansen at both places. This is not accusing people, this is pointing out procedural issues which need to be cleaned up.

        We also know that Lana Hires (who has now retired) took voter registration documents home. We know this because we found one discarded in her trash!"
    • Black Box Voting, "Devastating hack proven - Leon County dumps Diebold", 2005/12/13
      • "The Volusia County hack of 2000 where there were 16,022 negative votes for Gore and over 4,000 positive for Bush appears to be a version of the Hursti memory card hack. The similarities are striking."
      • "By the way, the 4,000 votes added to Bush were in addition to the negative 16,000 for Gore in Volusia, but that 4,000 votes took place in Brevard County -- ironically, Dr. Herbert Thompson's home county. He was surprised when I told him that. To expand on the irony, the Brevard County 4,000 votes was attributed to the central tabulator -- the component that Dr. Thompson hacked in Leon County. (Dr. Thompson's book, "The Mezonic Agenda -- Hacking the Presidency" was written before he learned about Diebold or saw the Central Tabulator. He was amazed to learn that some of the fictional methods to hack the vote in his book -- ie. password replacement by cut and paste of the hash code -- actually exist in the Diebold system.)"
      • "That being said, we do know at least some components were alleged to be the same, because ES&S filed a copyright infringement lawsuit against Global at one point. Both Sequoia and ES&S systems should absolutely be examined!

        Global Election Systems was run by a Howard Van Pelt, another Texan, until Urosevich took over. There are some very interesting features about the confluence of time when Global bought I-Mark Systems from Urosevich and others, then installed Urosevich as its head, and the ES&S acquisition of BRC, with Sequoia scooping up part of the dessert.

        Jim, Kathleen and I are working on sorting some of that out right now."
      • "Why do you assume the defects have anything to do with incompetence or "weak" code bases and systems? Those who have examined the code said that it is actually quite complex--appearing at first glance to be messy but actually quite sophisticated. Furthermore, they have said that it actually took a lot of skill to keep the specific open doors in place as the code went through various versions.

        There may well be sloppiness, incompetence and/or collusion elsewhere in the company, in the ITAs and among election officers--but I don't think it's necessarily due to a lack of intelligence or skill, and certainly not by the top programmers. Everything about the history indicates that these crucial security vulnerabilities were intentional, and that keeping them in place has been high on the company's priority list."
      • "You wrote: Another question. I vaguely recall a report that people could access the counting process from a separate location. Something about the votes being uploaded to .... a media agency, and that the people at that agency could in theory send information the other direction, back to the .... tabulator?

        I've heard this, but don't have a lot of confirmation on it. We have confirmed, and videotaped, that the votes are often transferred to a depot (also called by many other names). In Volusia County, most of the discrepancies trace back to the Daytona Speedway Depot. What happened in this location is under study right now."
      • "The software made by Diebold provides a combination of obvious back doors and subtle back doors. And let me make a prediction: When someone reverse engineers the code, they will find that supposedly clumsy software contains tripwires for reverse engineering, a feature requiring rather elegant sophistication. When interviewing sources inside Diebold, it is clear that various kinds of knowledge is closely compartmentalized. The more obvious back doors may be lower level people, but I believe there is one or two people at most who know more.

        As Hopsicker says, you can make some money selling elections equipment, but you can make a lot of money selling elections. You go where the opportunity leads you. These companies, like Global Election Systems which Diebold bought, were cash-strapped and in debt. They're going to take advantage of the opportunity they have, and it looks like that's just what they did when they brought an embezzler in to rewrite GEMS."
      • "I'm already seeing the Democrats struggle mightily to frame whatever does come out as a Republican ploy. However, the presidential tampering is just a toxic byproduct of the rotten core at the middle: The long-term corruption in a growing patchwork of local elections jurisdictions, the enabler for presidential tampering."
      • "There are some differences, using your theory. Diebold's selling point was that county election officials could program the removable media themselves. ES&S kept that procedure in house (rather horrifying, actually).

        Hart Intercivic also does that, from what I understand."
    • BBV forum discussion in January 2006 on "Various questions and answers"
      • "Tampered with: Volusia County 2004: The evidence for this is circumstantial, but I believe it will pass the test of "preponderance of the evidence." I'm not sure it's to the level of "beyond a reasonable doubt." The evidence is in the form of internal reports, unexplained need for extra memory cards, and poll tapes that contain incorrect dates, times, and occasionally, results. The Volusia County report will come out this month.

        We're not done with Volusia. The evidence there and the Hursti evidence have opened up several new lines of inquiry."
      • "In Volusia County, one employee was given a key to the ballot vault and 24-hour access, plus that employee had access to the warehouse. We found homemade ballots and lots of ballot masters in the warehouse trash."
      • "5). Why claim Diebold's ballot printing is not accounted? It appears that their ballot printing shop is just that, a print shop.

        Because we have financial documents gathered from two separate locations (Everett and McKinney) showing internal figures that estimate a whopping 25 percent spoilage rate. That was mind-boggling. In addition, we have not one but TWO sources from the inside ballot printing shop who have taken it upon themselves to express concerns about what happens to the extra ballots. And added to that, I confirmed with King County elections chief Dean Logan yesterday that the county does nothing to account for the extra ballots printed unless they are delivered. That leaves a security loophole. Now, when you have an elections division that keeps "finding" more ballots, that is not just an academic concern."
    • 2004 Volusia County FL precinct results
  • South Florida issues
  • Washington voting problems
  • Indiana voting problems
    • WISH-TV archive of election issues stories from 2004
    • WISH-TV, "ES&S Project Manager Resigns Over Voting Machine Problems", 2004/05/11: "Voting machine problems drove former ES&S project manager Wendy Orange to resign. She says ES&S knew it had a software problem and then tried to hide it. "I was faced with a moral and ethical dilemma and I felt the only thing that I could do was come forward and tell the Marion County clerk what had happened," said Orange. [...] In her letter of resignation, Orange said she found the corporate philosophy at ES&S to embody unethical and disreputable practices. She said she had "personally witnessed open discussions of potentially illegal procedures.""
  • Alaska election
  • Arizona election
  • Wisconsin election
  • Bob Nichols, "UPDATE: Voting Machines Count Backwards in Okla.", 2004/11/27 (Daily Kos archive, Tulsa Word printing of election night vote totals) - somehow, they printed a 2002 cockfighting referendum instead of the 2004 presidential results


CyberNET and Bay Point

Other curiosities