An Analysis of Electronic Vote Switching Incidents During the 2004 U.S. Presidential Election Reported to the Election Incidence Reporting System (EIRS)
Purpose The purpose of this analysis is to characterize reported incidents of “vote switching” by electronic voting machines that occurred during the 2004 Presidential election, by type of machine, state, county, and which of the two major party candidates benefited from the incident. It is hoped that this characterization will provide additional clues to what went wrong with the 2004 Presidential election, and that these clues will be helpful in targeting efforts to improve the integrity of future elections in the United States.
Methods Data source The source of all data for this project is reports to the Election Incidence Reporting System (EIRS): https://voteprotect.org/index.php?display=EIRMapNation&cat=ALL&search=&go=Appl y+filter&tab=ED04, developed by the National Election Data Archive Project. All reports to this System involve the U.S. national election of November, 2004. The EIRS database includes 28,734 reported incidents (as of mid-May, 2005, when the data was compiled for this study), including 2,115 “machine problem” incidents. The material for this analysis was obtained by searching these “machine problem” incidents only in counties that used electronic voting machines, according to a database provided by Voters Unite! (www.VotersUnite.Org ). Definitions A report was categorized as a presidential vote switching incident if and only if it met both of the following criteria: 1) The report specifically referred to the presidential vote (unless referring only to third party candidates) OR to one or both of the two major parties (unless referring specifically and only to non-presidential candidates), thus implying reference to the presidential vote; and 2) The report noted that the voting machine made it easier or more difficult to vote for one of the two major candidates OR difficult to vote for president in general. Typically these reports involved a voter attempting to register a vote for one candidate, and then the machine noting that another candidate had been selected. These “vote switches” involved switches from one to the other major party candidate, from a major party to a 3rd party candidate, or vice versa. Other problems involved such incidents as attempting to vote for a candidate and the vote not registering at all.
There were two exemptions from the above noted criteria: 1) Reports where the only problem noted was that one of the candidates was “pre-selected”. These were not included in this analysis because these cases likely involved a situation where the previous voted failed to register his/her vote, and the situation could be easily remedied by clearing the screen; and 2) Reports where I found it impossible to decipher what the complaint was referring to. Having determined that a report met the criteria for “vote switching”, the next step was to categorize the report into one of three categories, according to which candidate the incident apparently potentially benefited: Kerry, Bush, or neither. The incident was categorized as potentially benefiting Bush if it indicated either that: 1) The voter attempted to vote for Kerry (or Democratic Party), but the machine registered Bush (or Republican Party) or another candidate; 2) The voter attempted to vote for a third party candidate, but the machine registered Bush; or 3) The voter’s attempts to vote for Kerry (or Democratic Party) were made difficult by any other machine related activity. By interposing the words “Bush” and “Kerry” in the above noted criteria, we obtain the criteria for categorizing an incident as apparently potentially benefiting Kerry. And if the report failed to meet either of the above two criteria, then it was categorized as potentially benefiting neither candidate.
Assessment by swing state status A comparison of vote switching by swing state status assessed only the 87 cases of vote switches that favored Bush, since the number that favored Kerry was too small to analyze. The number of vote switches in both swing states and non-swing states was divided by the number of official votes from those states that were reported from counties that used electronic voting machines (rounded off to the nearest 1,000 for each county.) Swing states were considered to be CO, FL, IA, MN, MI, NH, NM, NV, OH, PA, and WI.
Results Overall categorization of reports Appendix A is a verbatim listing of all of the reports included or considered for inclusion in this analysis. Those highlighted in blue are those that were determined to apparently potentially favor Bush. Those highlighted in red are those that were determined to apparently potentially favor Kerry. Those highlighted in green are those that were determined to potentially favor neither Bush nor Kerry. And those highlighted in yellow were those that were excluded from this analysis on the basis of the exclusion criteria noted above. The following denotes how many reports were placed in each category: Apparently potentially favored Bush: 87 Apparently potentially favored Kerry: 7 Potentially favored neither: 52 Excluded from analysis: 8
Categorization of reports by state and county Table 1 shows the categorization of reports by state and county. Only counties with three or more reports in favor of one of the two candidates are specifically included in this table. Reports from counties with only one or two reports in these categories are included in the state totals, but not specifically listed. The 87 reported incidents that favored Bush, compared to the 7 that favored Kerry represent greater than a 12 to 1 ratio in favor of Bush for the U. S. as a whole. By far, the greatest excess of incidents favoring Bush occurred in Florida, where there were 47 reports favoring Bush (more than half of the total for the whole country), and only one favoring Kerry. Forty two (42) of these reports (89% of the total for Florida) came from three counties: Broward (23), Palm Beach (11), and Miami-Dade (8). Other states with a significant excess of incidents favoring Bush included Ohio (8 for Bush vs. 0 for Kerry), Georgia (6 for Bush vs. 0 for Kerry) and New Mexico (8 for Bush vs. 2 for Kerry). All of the incidents favoring Bush from Ohio were from Mahoning County, and all those from New Mexico were from Bernalillo County, whereas those from Georgia occurred in five different counties. Florida accounted for 54% of the incidents favoring Bush, and the other three states accounted for 25%, leaving only 18 incidents (21%) in the 46 remaining states.
Categorization of reports by voting machine vendor Four voting machine vendors accounted for all but three of the 87 reported incidents that were favorable to Bush. These included Diebold (7 incidents, 8%), Danaher (14 incidents, 16%), Sequoia (19 incidents, 22%), and ESS (44 incidents, 51%). Although these percents were very different than the distribution of voting machine vendors throughout the United States, all four of these vendors were characterized by a significant excess of incidents favorable to Bush, compared with incidents favorable to Kerry.
Comparison of swing states versus non-swing states Overall, of the 87 vote switches that favored Bush, 67 were reported from swing states, and 20 were reported from non-swing states. Counties from swing states that used electronic voting machines accounted for 10,104 K votes, and counties from non-swing states that used electronic voting accounted for 27,713 K votes. The rate of reports per million voters in swing states was therefore 6.6 per million, and the rate in non-swing states was 0.72 per million voters. Therefore, the rate of reports of electronic vote switching that favored Bush was more than 9 times greater in swing states than in non-swing states.
Assessment by rate of incidents per population voting on electronic machines The distribution of incidents by state and county was very uneven. Of the 67 vote switches reported from swing states, all were reported from four states: FL: 47 incidents, rate = 11.5 per million OH: 8 incidents, rate = 9.1 per million NM: 8 incidents, rate = 11.7 per million PA: 4 incidents, rate = 2.6 per million Three swing states, WI, NH, and MN, had no reports because they did not use electronic voting machines. And the other four, IA, NV, MI, and CO, simply had no reports even though some counties in those states did use electronic voting machines. Of the 24 non-swing states that used electronic voting machines in some counties, only Washington exhibited a rate of reports that equaled any of the swing states noted above. Washington State had 3 reports, all from Snohomish County, for a rate of 8.2 per million for the state. Distribution of incidents also varied greatly by county. Five swing state counties accounted for 58 of the 67 swing state reports (87%) and much higher rates than any of the other counties: Broward, FL: 23 incidents, rate = 32.5 per million Miami-Dade, FL: 8 incidents, rate = 10.3 per million Palm Beach, FL: 11 incidents, rate = 20.1 per million Bernalillo, NM: 8 incidents, rate = 31.4 per million Mahoning, OH: 8 incidents, rate = 60.2 per million
Discussion The most striking finding of this analysis was that there were far more reports of electronic voting machine switches favorable to Bush, compared to incidents favorable to Kerry, with a ratio of greater than 12 to 1. The other major finding was that of the 87 reported incidents that favored Bush, these reports were characterized by more than 9 times the reporting rate (per voter using electronic machines) in swing states than in nonswing states. This poses the question: What is the meaning of these findings? As with any study which demonstrates an uneven distribution of a variable, there are three possible alternatives: chance, bias, or causal connection. Let’s examine each of these possibilities:
Chance The likelihood of such an uneven distribution of Bush vs. Kerry favorable incidents is similar to the likelihood of flipping a coin 94 times and coming up with 87 or more heads or tails. The odds against that exceed ten million to one.
Bias Bias would explain the uneven distribution if in reality the Bush and Kerry favorable incidents occurred with approximately the same frequency, but the Bush favorable incidents were more likely to be reported. This possibility is similar to the hypothesis posed by Warren Mitofsky to explain the discrepancy between his November 2004 Presidential exit poll (which had Kerry winning by 3 %) and the official election results (which had Bush winning by 2.5%). Mitofsky’s hypothesis is that Bush voters were less likely to participate in the exit poll than Kerry voters. Similarly, the findings of this analysis could be explained if Bush voters were much less likely than Kerry voters to report incidents where an electronic voting machine appeared to make it difficult to vote for their candidate or appeared to switch their vote to the other major party candidate. Although I don’t find it difficult to believe that such a bias could exist, I do find it very difficult to believe that the magnitude of such a bias could be so great as to result in a 12 to one ratio. I can’t say it’s not possible. But it seems to me to be a very unlikely explanation.
Causal connection The remaining possibility is that there were many voting machines throughout the country with which it was more difficult to vote for Kerry than for Bush, or which switched or attempted to switch votes from Kerry to Bush, and that these machines were concentrated in certain areas of the country: Southeast Florida, Mahoning County, Ohio, and Bernalillo County, New Mexico. This could have been accidental. But if it was accidental, then why would the vast majority of these incidents tend to favor one candidate over the other? I cannot think of an explanation for this, but perhaps someone with expertise in the computer programming of voting machines could. If the tendency of these voting machines to favor Bush was not accidental, that means that someone programmed them to act this way. Depending on the magnitude of this phenomenon, that could have compromised the integrity of the election. This is especially true given the fact that Florida and Ohio were the two states where this problem was reported with the greatest frequency, and the fact that if either if these states had gone for Kerry, he would have won the election. There has been much speculation and accumulated evidence of compromised integrity of the 2004 Presidential election. Much of this has centered on the fact that the MitofskyEdison exit polls not only showed Kerry winning the national vote by 3%, but also showed Kerry winning Ohio by 4.2% (which official results showed him losing by 2.5%) and virtually tied in Florida (where official results showed him losing by 5.0%): http://electionarchive.org/ucvAnalysis/US/Exit_Polls_2004_Edison-Mitofsky.pdf . In addition to exit poll discrepancy evidence, there has been a good deal of more direct evidence accumulated of compromised election integrity, such as the Rep. John Conyers report by the Democratic staff of the House Judiciary Committee: Preserving Democracy: What Went Wrong in Ohio: http://www.truthout.org/docs_05/010605Y.shtml.
Evidence more directly pertinent to this analysis is testimony given before the House Judiciary Committee by an ex- Florida computer programmer, Clint Curtis: http://www.rawstory.com/images/pdfs/CC_Affidavit_120604.pdf. Curtis testified that he was requested in 2000 by Tom Feeney, then Speaker of the Florida House of Representatives, to “develop a prototype of a voting program that could alter the vote tabulation in an election and be undetectable”. He did develop the program, after telling Feeney, however, that he could not make the program so that it would be undetectable if the source program were to be inspected. What is the magnitude and significance of the problem – is this the tip of an iceberg? If voting machines used in the 2004 Presidential election were in fact programmed to make it more difficult to vote for Kerry than for Bush, or to switch votes from Kerry to Bush, what significance could that have had to the integrity or outcome of the election? 87 individual incidents in an election where Kerry lost Ohio by over a hundred thousand votes and Florida by a few hundred thousand votes doesn’t seem like very much. But what if these 87 incidents represent only the tip of an iceberg – the known part of a much larger problem? There are many reasons to believe that these 87 reported incidents represent only the tip of the iceberg. These reasons include: Perhaps only a minute fraction of problems discovered by voters were reported to EIRS No one knows what fraction of problems discovered by voters at the polls in November 2004 were reported to the EIRS. It could be that the great majority of voters weren’t even aware that the system existed. Or if they were aware of it, they may not have felt the necessity of taking the time to report it. Most voters may not have noticed the problem A typical report noted that a voter would attempt to register a choice for President (or other candidate), and then prior to finalizing their choice would note that the screen registered a vote for the other candidate. How many voters would have noticed this, and how many voters would have failed to notice it, and therefore cast their vote for the other candidate? What if vote switching was usually not accompanied by any visual evidence? If the machines were programmed to switch votes, the person(s) behind this crime would not have wanted the machines to register any visual evidence to that effect, thus enabling the voter to have a chance to correct the problem, or bring it to the attention of election officials, and potentially a much wider audience. But what if it was not possible to program the machines in such a way that they wouldn’t occasionally provide this evidence, or what if doing so would have required a level of skill that many of the programmers didn’t have? In that case, the great majority of vote switching would have gone unnoticed and uncorrected – and therefore unreported. This is speculation on my
part, since I do not have the computer expertise to know how feasible such a scenario would be. Many of the individual reports note that the problem had been occurring all election day long. Analysis of Snohomish County, Washington A report by Paul Lehto and Jeffrey Hoffman which identifies 19 reports of electronic vote switching in Snohomish County, Washington – all which favor Bush – from the Washington State auditor’s office, the Washington Secretary of State’s office, and a Snohomish County voter complaint hotline. This compares with only three reports made to EIRS. Washington Post Investigation in Mahoning County, Ohio This investigation identified 25 electronic voting machines in Youngstown, Mahoning County, which transferred an unknown number of votes from Kerry to Bush: http://www.washingtonpost.com/wp-dyn/articles/A64737-2004Dec14_3.html. The Post report goes on to state “Due to lack of cooperation from Secretary of State Blackwell, we have not been able to ascertain the number of votes that were impacted or whether the machines malfunctioned due to intentional manipulation or error.”
Perspective and conclusion With John Kerry now considering withdrawing from the lawsuit to force a full and fair recount of the Ohio vote, it is especially critical that all evidence of election fraud be thoroughly considered. This study adds one more piece to the accumulated evidence of compromised integrity of the 2004 Presidential election. The mystery of the 12 to one ratio of reported vote switching incidents that favored Bush compared to those that favored Kerry remains unexplained. Simple statistical tests show that this ratio could not have occurred by chance. The possibility of a massively large reporting bias seems unlikely, although it cannot completely be ruled out. The finding of a 9 to 1 ratio of these incidents in swing states makes the possibility of reporting bias as an explanation even less likely. It seems most likely, therefore, that the computers were programmed to make this happen – either accidentally or purposely. In either case, we know neither the magnitude of the problem, nor to what extent it might have affected the 2004 Presidential election results. A better understanding of this problem can be obtained only by inspection of the implicated computers. Since this problem may have affected the election results, and since if that did happen, the problem may be repeated in future elections, it would seem to be of the utmost importance that the implicated voting machines be examined by qualified experts. To my knowledge, that has not yet been done. Hopefully this report could help target the voting machines that need to be examined.