BY ELIZABETH JORDAN AND OLIVER TERENCE DAWSHED - The present work discusses how to identify electoral fraud and estimate its scale. Reasons to believe it occurred in Florida in 2000 and, with less certainty, in 2004 are outlined. It should be noted, however, that statistical analysis on its own is not sufficient to prove fraud. Proof can only result from direct evidence gathered by a searching investigation. A political strategy must be combined with an analytic strategy to obtain that investigation.

A critical element of that strategy is to develop a genuinely impartial media. Even if it were shown that votes were not being stolen, media performance has been so defective that American elections cannot be said to be free and fair. Indeed, two former American presidents and several senior members of Congress have indicated that the elections of 2000 and 2004 were not.

When election results come in, how can one know whether they are accurate or not? Computer voting may facilitate fraud, but even paper ballots can be corrupted, as the recent voting scandal in the UK [1] illustrates. A decision to request a recount must often be made within hours of poll closing, based on limited information, and with supporting evidence strong enough to persuade a judge. Political variables often determine whether recounts are obtained: complaints that are not timely or that represent minor or accidental wrongs are often dismissed as sour grapes.

With traditional voting, one can divide strategies of electoral fraud into several classes (electronic balloting eliminates the ballot, and therefore limits the conduct of an audit, but doesn’t substantively change the analysis):

• Illegal voting. An unqualified voter may register or a qualified voter may vote more than once. These are examples of illegal voting.

• Vote suppression. Officials can suppress voter turnout by applying arcane rules to registration or voting, changing polling places without notice, or assigning too few machines to a precinct. Agents of a candidate, whether official or freelance, can destroy voter registrations, threaten prospective voters, or spread disinformation about voting time or place in their efforts to reduce turnout.

• Ballot tampering. Officials and election workers have a rich smorgasbord of methods with which to tamper with ballots. They can destroy or lose ballots, void votes for certain races (for example, by pre-marking a ballot), or substitute fabricated ballots for genuine ballots. Providing ballots that are so poorly printed that voters are unable to correctly select the candidate of choice can also be considered a form of ballot tampering.

• Defeating the integrity of the counting machine. Election officials and technicians have the opportunity to tamper with machines used to count ballots. The proper functioning of counting machines requires that pens used on Mark-sense (optical scan) ballots have the right ink, for example. One strategy to corrupt Mark-sense readers [2] might be to align them so that the mark for a certain candidate is not centered where the optical beam will pass. Reading of the ballot will intermittently fail, decreasing the votes for that candidate.

• Vote count alteration. A limited number of county officials have access to final tabulations and can therefore manipulate the reported results. With the advent of electronic transmission of election results and computerized databases to store and process election information, the potential for tampering is almost unlimited.

Some of these strategies, of course, can overlap. For example, recent work by Richard Hayes Phillips [3] and Joe Knapp [4] illustrates how the assignment of too few machines to a precinct can both depress the total number of votes (voter suppression) and lead to an increased number of voter errors - which, if employed deliberately, amounts to ballot tampering. The point of differentiating between these strategies is to identify who might have the necessary access to commit fraud and how, in detail, it might have been done.

Illegal voting, which can and does lead to prosecution, does occur. Republicans have complained that Democrats encourage non-citizens or felons disbarred from voting to register and vote. Similarly, Republicans have charged that northerners with southern vacation homes vote in both states, a phenomenon called “snowbird” voting. There is no substantiation of Republican charges that illegal voting is widespread, however. Despite lurid allegations, prosecutions are rare. There is not a shred of evidence to indicate that illegal voting is condoned or promoted by Democrats. For example, a bit of thought might suggest that voters with two homes and the facility to travel between them (“snowbirds”) are likely to vote conservative, whatever their party registration may be.

Unfortunately, very little thought goes into analyzing allegations of illegal voting in the mainstream press. The same newspapers that dismiss “Internet conspiracy theories” eagerly dispense truly wild conspiracy theories about mass illegal voting by felons, immigrants, or derelicts. The fact that these allegations invariably dissipate on impact with reality does not seem to reduce the press’s enthusiasm for throwing them out.

Vote suppression is a strategy widely employed by the Republican Party. It is usually deployed primarily against minority voters, which may be why it is very rarely punished. The case of Sproul & Associates [5] is typical. This company, which was paid over a hundred thousand dollars by the Republican Party [6] to gather voter registrations in a number of states, is alleged to have systematically blocked prospective Democrats from registering and destroyed numerous Democratic registration cards. Given the number of separate specific allegations, it is difficult to believe that there was no substance to them, yet courts and the Federal Election Commission have apparently taken no action. Similarly, no one has been punished for the well-documented directives by the Florida Department of Elections ordering county election officials to remove tens of thousands of legal (and disproportionately African American) voters from the rolls in Florida [7] just before the 2000 election.

Tampering with ballots or with counting machines and reporting false election results are criminal acts. An allegation of fraud by these means is serious indeed, and should not be made lightly. The evidence is sufficient to justify a searching investigation of Florida’s elections, though it is not sufficient to sustain an allegation of wrongdoing by any specific individual.

For the most part, charges of fraud by tampering with ballots or counting machines and reporting false results are often regarded as implausible since, at least for a statewide or national race, it is presumed that many people would have to conspire to achieve it. The assumption that a large conspiracy would be needed to steal an election is certainly wrong, in the sense that in elections getting more votes than the opponent is a goal so clear that no overt conspiracy is required. When the goal is widely known, “leaderless resistance” [8] is a common strategy, with a long history among those who oppose equal rights. It would not be surprising if, thanks to residual racism, the worst voting machines ended up in African American precincts. Whether that would amount to electoral fraud or just widespread individual vindictiveness can be debated, but the effect is the same: people are denied the right to choose their leaders.

Furthermore, the Iran-Contra scandal, in which weapons were illegally transferred to an enemy of the United States (Iran) and to terrorists in Central America, shows that conspiracies involving hundreds or thousands of individuals can go on for years without media notice. Iran-Contra might well still be secret had not a freak plane crash brought to light US government involvement. Of course, thanks to the fact that nowadays the publication of probing investigative reporting is just as freakish as light plane crashes, obtaining the political leverage necessary to force timely recounts is hard to come by in presidential races. Without proper auditing, almost anything could happen.

There are very few reliable data with which to analyze the results of an election. Statewide, there are exit polls for major candidates, but getting the raw data by precinct is not so easy. The vote count by candidate in each precinct is usually available, though sometimes counties refuse to provide even that data, releasing only countywide totals. The number of registered voters in each precinct is usually known.

Sometimes other information, such as the number of voting machines in each precinct, is available. Phillips and Knapp applied these data to the 2004 Ohio elections to construct a persuasive case for voter suppression. Recently, Paul Lukasiak showed that the frequency of certain kinds of overvotes can be indicative of fraud. [9] Census data broken down by precinct has become available, allowing one to derive the percentage of adults who are registered to vote and other items of potential interest.

Sometimes one can obtain the number of miscast or spoiled votes, either directly or as the difference between the number of voters who showed up at the polls and the total number of votes recorded in a given race. More rarely, one can obtain a breakdown of spoiled votes as overvotes, in which more than one candidate was marked in a given race; or undervotes, in which no vote was read for any candidate. Often there are discrepancies between the sign-in book and the vote count. If there are more votes cast than known voters, the result is phantom votes. Ellen Theisen and Warren Stewart identified over a thousand phantom votes in the 2004 results from New Mexico.

The frustrating reality is that, at least in many states, one is at the mercy of county officials, who choose to release or withhold data as they will, and provide it either in a convenient format such as a spreadsheet or in a format that requires tedious reprocessing or transcription, such as PDF. So, in devising auditing tools, keeping it simple and easy is critical. Fortunately, most strategies of tampering with ballots or counting machines are potentially detectable by simple statistical methods.

The simplest method is to examine the number of spoiled ballots by precinct. For precincts that use the same voting machine type and ballot design, the fraction of spoiled ballots should be nearly the same. When spoilage in certain precincts is higher than in others, either there is a concealed difference in voting machine type or ballot design, or there may have been a violation of equal protection under the law. Since the Voting Rights Act [10] requires that disability not be a factor in access to voting, the age, eyesight, or intelligence of voters is not an excuse for unusual ballot spoilage. Race, too, is excluded as a legitimate excuse. Defenders of the status quo may fall back on education as an excuse, but there is very little evidence that education affects the ability to vote.

Professor Allan Lichtman [11] developed the basic technique used to determine discriminatory effect. Called ecological regression [12], this technique as applied here amounts to graphing the amount of spoilage versus the percentage of a minority population in a voting district and drawing the best-fit straight line through the data points. One can refine this procedure in a number of ways, such as weighting the voting district by size.

In “Bush’s Fifth Ace: a Crooked Panhandle,” [13] we demonstrated that ballot or machine tampering strategies have predictable effects on electoral outcomes. One of the most interesting and generally useful of these was the effect on crossover. In crossover voting, a voter who votes Democratic for one race decides to vote Republican for another or vice versa.

In Florida 2000, a study of 3 million ballots by Dan Keating of The Washington Post [14] (11/12/01) concluded that crossover was, on average, only 20% for Republicans and slightly less than that for Democrats in the race for US Senate. In a precinct in which 100% of the voters voted for Gore, 20% might vote for Republican Bill McCollum for Senate. In a precinct in which 100% of voters voted for George Bush, 20% might vote for Democrat Bill Nelson for Senate. Even those extreme values overstate the situation, since very few precincts voted more than 80% for any presidential candidate.

On the other hand, there could be some regional variation in crossover. One might expect Bill McCollum to be more popular near the congressional district he represented in northeast Florida, while Bill Nelson might be more popular in Tallahassee where he served, or in central Florida where he grew up and served at NASA. Still, by looking at precincts rather than counties, one should get around the issue of regional bias.

What we found was striking. In certain precincts, there was crossover well outside the norm. In heavily Republican precincts, an unusual number of voters who voted for Democratic Senate nominee Bill Nelson also voted for George Bush. Additionally, we found that the counties in which we found this unusual sort of crossover had very high levels of spoilage in Democratic precincts. Combining the two factors, which we proved were independent of one another, we were able to explain almost all of Gore’s performance. The patterns were so regular that they did not resemble what is seen in nature. They looked systematic, as if produced by machine. We called this strange crossover “pseudocrossover” to reflect our opinion that it was probably the result of tampering with ballots, counting machines, or final tallies.

The value of this work was that it identified a small number of precincts that were the most likely locations of electoral fraud and estimated the scale of fraud. In 2000, we estimated that 7100 votes could have been stolen in 11 northern counties and more votes in other counties. This is not massive, but it certainly was enough to swing the election.

In more recent work, we have shown that

• spoilage was all but eliminated in 2004, but pseudocrossover was not,
• pseudocrossover cannot be explained by simple models of crossover,
• pseudocrossover cannot be explained by “Dixiecrat voting” (whatever that is),
and

• there is an atmosphere of official lawlessness in Florida consistent with fraud.

Scholars and lay researchers have developed effective strategies for identifying electoral fraud. To effect change, however, requires both a political as well as an analytic strategy. Efforts to force a serious investigation of the Florida vote have failed so far because Floridians have not been willing to do the on-the-ground investigation needed to translate statistical analysis into information useful in a criminal investigation. Thanks to The Free Press [15] and The Alliance for Democracy, [16] Ohioans in 2004 have made more progress. The battle, though, is far from won.

If elections are being stolen - a suspicion that must be regarded as not yet proven - the theft will only stop when someone pays a heavy price. If elections are not being stolen, then a proper investigation might reestablish some degree of faith in our institutions. It is not rational to wait for a favorable election cycle for change to be ushered in. Faith in fair elections is essential to a free country, since without confidence that votes will be counted, voters have no reason to go to the polls.

In a recent Newtopia article, Diane Perlman suggested a number of reasons why voters may be apathetic [17] in the face of the possibility that elections are being stolen. If they can be involved in efforts to monitor elections, that could change. Certainly the Florida effort we have mounted needs volunteers for specific tasks. Many of these, such as building database queries, contacting counties to request vote information, or collating information available online can be done over the Internet.