FAQ
By Oliver Terence Dawshed
Revision 2.0   4/16/05
Subject to revision. Corrections are welcomed.

1.  What are your findings? 

Methods for detecting electoral fraud from electoral results.  The most important accomplishment of our work was to show how one would go about detecting electoral fraud. We went through, step by step, to show exactly how tampering with conventional paper ballots would show up in electoral results. Based on simple logic, we developed statistical methods to analyze voting patterns. Briefly, we found that two items (spoilage and crossover) were sufficient to describe most forms of ballot tampering. The advent of electronic voting facilitates tampering, but the same principles hold.

Spoilage in 2000 election hit Democrats.  For the 2000 election, we concluded that the votes cast but not counted (so-called "spoiled" votes) were disproportionately those of Democrats. We accepted the conclusion of the US Commission on Civil Rights (USCCR) that spoiled ballots were disproportionately those of African Americans. African Americans are disproportionately Democrats. Our results therefore mesh with those of the USCCR.

A second factor, pseudocrossover, defined.  We further concluded that there was a second important statistical anomaly in the 2000 election, one that superficially resembles crossover voting, but is not fully consistent with real crossover. Real crossover is seen as a difference between the number of votes for one party's president and the vote for that party's senator.  For example, in Florida 2004, the crossover ratio is Kerry votes/Nelson votes.  What was observed in Florida differed from real crossover because in 2000:

•     it was not explainable by regional candidate preferences,
  
•     it was not explainable by simple models of crossover,
  
•     it varied by precinct in a manner that was closely correlated to partisan outcome,
  
•     it seemed to occur in the same counties that partisan-targeted spoilage occurred,
  
•     it was almost as powerful an explanatory factor as spoilage, and
  
•     alternative explanations, notably the "Dixiecrat hypothesis," failed.

We called this second factor pseudocrossover. Between pseudocrossover and spoilage, we were able to explain the majority of Al Gore's electoral performance in almost all of the counties we looked at. In several counties, there was almost no random variation!  

2004 elections confirmed the observations.  For the 2004 election, better electoral practices in Florida drastically reduced spoilage, but there were still counties in which spoilage appeared to have occurred disproportionately in Democratic precincts. Pseudocrossover remained an important factor. However, the patterns of individual counties appeared to be substantively different from 2000, consistent with the notion that the crossover voting observed is an artificial result.

2.    Were the elections of 2000 and 2004 stolen?

Two presidents and senior members of Congress say yes.   We do not need statistical evidence to reach the conclusion that the 2000 and 2004 elections were not free and fair. Indeed, former President Jimmy Carter said that the 2000 election in Florida did not meet the basic standards for a free and fair election.[2.1] Bill Clinton has repeatedly stated that he believed Al Gore won Florida.[2.2] In 2004, a formal inquiry led by Congressman John Conyers, Ranking Member of the House Judiciary committee found abuses in Ohio in 2004 that were so serious that the committee could not state with any certainty that the president had been lawfully elected.[2.3] For only the second time in the nation's history, in 2004, Congress challenged the legitimacy of a presidential election. 

Free and fair elections require fair reportorial, electoral, and judicial process.  The problems were much deeper than what Carter, Clinton, and Conyers were willing to concede. The news media grossly misreported basic facts about the Democratic candidates and suppressed information adverse to the candidacy of George W. Bush. We know that senior Florida Republicans, in flagrant violation of a court order, mandated the removal of a large number of legal (and mostly Democratic) voters from the voting rolls. Legal scholars, including a number of conservatives, have stated that the judicial process used by the Supreme Court to intervene in the election was itself lawless.[2.4, 2.5, 2.6] If the events of Elections of 2000 and 2004 had occurred in any other country, we would have called that country a banana republic, a dictatorship. 

Genuinely democratic nations vigorously defend free elections.  The Founding Fathers would have regarded any one of these abuses as intolerable. They would have denounced any nation that did tolerate them as corrupt beyond redemption. Voting is a right that is prior to all others. If elections are not free and fair, or even if a populace comes to wrongly believe that they were not free and fair, the consent of the governed is absent and the rule of the government rests on sand.  The failure of most of our electoral officials to defend the principle of a free election in all its dimensions speaks of a political process that has rotted out. If leaders have forgotten what a country is about, why should average citizens remember? 

3.   Do you allege that there was electoral fraud in Florida in 2000 and/or 2004?

No.  Not to be fussy, but words have meanings. "Allege" does not mean "think" or "believe." Certainly, based on a great deal of consideration and analysis, we believe that there was electoral fraud. However, to make an allegation of fraud, one must prove (or attempt to prove) the commission of a specific crime. To illustrate the limitations of statistics, consider that they can show that the DNA of a murder suspect matches that of material found at the crime scene to a given level of certainty. But if it were found that fingerprints on the murder weapon were not those of the suspect, the DNA evidence would have no value in the courtroom, no matter how high the certainty of a match.  

Electoral statistics can show, to a certain level of certainty, that the voters of a certain district did not vote in the way that was officially reported. But statistics give very little help in figuring out who altered the vote or how it was accomplished. One would need evidence of "who?" and "how?" to make an allegation.

To take an obvious example, disproportionate ballot spoilage in African American precincts almost certainly represents a violation of law. But was the means by which it was accomplished electoral fraud? Or was it simply a violation of the Voting Rights Act?  If ballots were deliberately altered, electoral fraud was committed.  But if the least well-maintained voting machines were sent to African American districts through bias rather than conscious intent, it might not rise to the level of electoral fraud. And in theory, it's always possible that the voters deliberately spoiled the ballots, in which case no crime at all would have been committed. So, statistical analysis is not sufficient on its own to make an allegation of electoral fraud. Which individual would we accuse?  Of which crime would we accuse them? 

What our work accomplishes is demonstrating that the results probably did not occur by chance and that they are consistent with electoral fraud. The analysis points to a relatively small number of precincts where any fraud would most likely be found - if there were a proper investigation. We have called for such an investigation, but have not pre-judged what it would find. 

4.   Why did Dawshed and Jordan conclude that there was probably electoral fraud in Florida 2000?

Very simply, because electoral fraud was the most plausible explanation of the many we considered for the observations of Florida's 2000 election. Statistical reasoning is a key element of deciding what is plausible and what is not, but it is only one key element of several. In New Hampshire, 2004, for example, election officials - though obviously not eager to host a recount - showed simple respect for the law. They did not attempt to obstruct reasonable steps to recount. That behavior seemed to suggest that they had nothing to hide. Had Florida behaved with equal grace in 2000, there would not be a cloud over George Bush's election. We also know that Al Gore would have almost certainly been declared the winner. 

Statistical reasoning.  Statistical analysis was one step in concluding that there was probably electoral fraud. Statistics showed to a very high level of certainty that:

• The spoilage of ballots was about ten times higher in African American precincts than in white or Hispanic precincts, independent of voting machine type.
  

• There was an unusual pattern of voting, called pseudocrossover, in many Florida precincts.  The voting patterns for President and for US Senator were so inconsistent that the result was almost certainly not by chance. No plausible explanation other than fraud has been presented.
  

• The two preceding observations are independent of one another, yet occur in the same counties in a striking number of instances.
  

• Pseudocrossover cannot be explained by simple models of crossover voting. Very complex models have to be assumed, and it has to be assumed that voter behavior changes radically from election to election.

Context supplements statistics.  At some point, reason has to take over from statistics. If we find a cabinet in disarray, we wonder what has happened. If, in another house, we find a smashed window, we wonder what has happened. But when the two observations pertain to one house, indeed to our own home, we start to suspect that a crime has occurred. Just so, the finding of unexplainable crossover and targeted spoilage provides context to suggest that neither is accidental.

The political context of Florida 2000.  In Florida 2000, there were plenty of warning signs of fraud.

A.  There was an atmosphere of lawlessness in Florida, and it came from the top. 

• A prominent Republican, Miami Mayor Xavier Suarez, was implicated in a vote buying scandal, in which a top aide (Humberto Hernandez) and 56 other people were convicted. The Republican Party put him in charge of its absentee vote effort.[4.1, 4.2, 4.3, 4.4] In another absentee vote scandal, Republican activist Isis Segarra was accused of manipulating disabled, elderly, and non-English speaking voters. She had been convicted of electoral fraud in 1983, but the conviction was overturned.[4.5]

• As is well known, the Secretary of State's office ordered a purging of the voter rolls just before the election.[4.6] What is not so widely known is that the manner in which the purge was done, which wrongly removed primarily African Americans from the voter rolls, was a flagrant, in-your-face violation of existing court orders.[4.7, 4.8]

• Ignoring basic ethical principles, the senior election official, Secretary of State Katherine Harris chose to also co-chair the Bush-Cheney campaign, served as a delegate at the Republican convention, and campaigned along with Governor Jeb Bush on behalf of George W. Bush.[4.9]

• Governor Jeb Bush sent out a letter under state seal urging that voters vote absentee despite the fact that Florida law forbids the use of the state seal for partisan purposes and requires that absentee voting be done only by necessity and not for convenience. This was challenged ahead of the election. The judge, Terry Lewis, said that there should have been criminal prosecution for misuse of the seal. In Bay County, Judge L. Ralph Smith, Jr. said there should have been criminal prosecution of misuse of the seal and abuse of absentee ballot.[4.10]

• Officials in Escambia and Manatee Counties admit to having deliberately disabled overvote protection. However, strangely, it functioned in Republican precincts, at least in Escambia. In Orange and Columbia Counties, overvote protection was overridden manually - but not in Republican precincts. One precinct in Bay County proposed initially that a power failure had invalidated ballots 15 percent of ballots in one precinct. When that explanation failed to hold up, they blamed the use of improper pens.  That precinct was 67% Democratic.[4.11] Escambia, despite its claimed frugality in forbidding voters a chance to correct errors, duplicated 2,400 absentee ballots - 11% of all ballots that were similarly treated in the state - to make them machine readable. Documentation of this practice was so loose that no one could say how many were duplicated because of problems with the presidential race.[4.12]

• In Bay County, 1,478 ballots were not machine readable and were duplicated. In flagrant disregard of the Sunshine Law, officials refused to permit inspection of the original ballots.[4.12]

• The Commission on Civil Rights concluded that there were violations of Section 2 of the Voting Rights Act, specifically a denial of voting rights.[4.13] The Ashcroft Justice Department conceded that there had been disparate treatment of minority and disabled voters.[4.14]

• Polling places were unlawfully moved without notice or closed early.[4.13]

• Florida state troopers placed at least one unauthorized police checkpoint in what clearly appeared to be an attempt at voter intimidation.[4.13]

• The Governor's Office of Executive Clemency, responsible for restoring voting rights taken away from felons convicted in Florida unlawfully attempted to force felons whose voting rights had never been taken away to apply for clemency.[4.13]

• Remediation of absentee ballot applications done in Martin and Seminole Counties was almost certainly illegal. The judge for a related civil case in Seminole County said that the County Clerk should have been prosecuted, not sued in civil court, for her role in this. At least in Seminole County, where remediation was done in the county clerk's office, security of voting records for the entire county was compromised.[4.10, 4.15, 4.16]

• Nassau County did an automatic recount, as required by law, but when this resulted in 51 more votes for Al Gore, the Republican commissioners unlawfully reported the original result. They simultaneously, and unlawfully, added in a ballot that had not been previously counted, a military ballot with no date or postmark.[4.17]

• Republican Party operatives, on the payroll of the US government, staged an unlawful demonstration for the purpose of interfering with the lawful execution of duties by Miami officials. They are alleged by Luis Rosero to have committed at least one act of violence. [4.18, 4.19, 4.20]

• A computer programmer, Clint Curtis, filed an affidavit alleging that a senior Republican official, then Florida Representative and now US Representative Tom Feeney, commissioned the production of a prototype of software to steal votes in September 2000.[4.21] It is not known whether the effort went beyond prototyping. 

• Republicans insisted on the counting of 680 overseas ballots which violated state law in lacking postmarks, having been mailed after the election, having been mailed from within the US, having been Faxed in, or lacking witness signatures. Nineteen ballots were from voters who had already voted! Seventeen percent of the ballots were not postmarked, vs. 1% of military mail that normally misses a postmark.[4.22, 4.23]

• Congressman Steven Buyer and other House Republicans requested telephone numbers or e-mail addresses for members of the armed service whose ballots had been disqualified, in an attempt to use them in a public relations efforts on behalf of the Bush campaign. This was a use of taxpayer dollars for political efforts and a violation of the requirement that military personnel stay free of politics.[4.24]

• Republicans in the Florida legislature refused to accede to the rule of law, planning to legislatively direct electors to vote for Mr. Bush, no matter what the real outcome of the election or the finding of the court. This was plainly unconstitutional, since the Constitution provides that a dispute over electors is to be resolved by the House of Representatives. Governor Jeb Bush also acted with unseemly haste to file the certificate of ascertainment that formally delivers the slate of electors to the federal government. [4.25]

• Four Escambia County Commissioners were later indicted on charges such as bribery, racketeering, and money laundering in relation to land deals. While crooked land dealing doesn't imply crooked vote stealing, it does reflect poorly on Panhandle sensibilities of morality.[4.26, 4.27, 4.28]

• In our own attempts to get data, we encountered a shocking disregard for Florida's Sunshine Laws. Counties spent months replying to simple requests or simply refused to respond.

• I filed a Freedom of Information Act request with the Pentagon to get information about military ballots. The disrespect for lawful procedure I experienced was astonishing, though apparently all too typical.

  
  

B.  The Republican Party behaved as if there had been fraud. Specifically:

• Representatives of George Bush filed actions in court to block any recount, acted to stall lawful recounts with frivolous challenges - at least 15,000 in Palm Beach alone - and, as described above, unlawfully interrupted recounts in progress.[4.29, 4.30] In addition, Jeb Bush's acting general counsel, Frank Jimenez, intervened to cause the head of the Division of Elections, Clay Roberts, to rule that hand recounts were illegal, even as the George W. Bush campaign had hand recounts yielding 418 Bush votes done in 6 northern counties.[4.31, 4.32]

• Katherine Harris allowed Republican partisans to work in her office, may have attempted to unlawfully erase files from her computer's hard drive, and deleted e-mails to and from Governor Jeb Bush almost certainly to cover up a misuse of state property and time for partisan purposes. As additional evidence of her indifference to the law, she also had prohibited political materials on the computer, had written a November 26^th, 2000 speech showing she had pre-judged the Palm Beach recount, and was even unlawfully registered to vote in two counties (willfully voting in the wrong place in Florida could cost one five years in jail, as one Edwin McGusty discovered to his dismay). Jeb Bush's office made 100 calls to his brother's campaign at taxpayer expense; even though 10 were made by Jeb Bush himself, he denied knowing what they were about.[4.33, 4.34, 4.35, 4.36, 4.37, 4.38, 4.39] The unmistakable impression was of a state government filled with officials who lied as easily as they breathed, whose notion of the law was that it was a convenience to advance their wants or deny those of others.

• John Stafford, supervisor of Duval County, is said to have misled the chairman of the Gore campaign in northeast Florida, Mike Langton, about the massive number of spoiled votes in Duval County, presumably to prevent Langton from  filing a timely protest of the Duval results.[4.40] In violation of Florida's Sunshine Laws, Stafford refused to permit the inspection of Duval County ballots and had to be taken to court to force the disclosure.[4.41, 4.42]

• Republican appointees of the Supreme Court contradicted their own earlier finding in Bush v. Gore and ignored the actual arguments that were presented in session before them to declare that recounting of ballots would be a violation of equal protection before the law.  This ruling has been widely critiqued by legal scholars, including conservatives, as an unprecedented case of judicial activism amounting to lawlessness, even according to at least one of the Court's members, Justice Steven Breyer, who called the decision "outrageous" and "indefensible."[4.43, 4.44, 4.45]

• Following the exposure of the deliberately fault voter purge scheme - one that flagrantly violated court orders consented to by the state - the Republican-controlled Florida legislature apparently attempted in August of 2001 to continue the illegal purge through a budget amendment. Though downplayed by the media at the time, in fact the Division of Elections went ahead with a new defective purge, lending credence to suspicions voiced at the time by Bob Fertik of Democrats.com that the Legislature was trying to secretly proceed with purging. The state abandoned the attempt only when confronted with the fact that the list contained no Hispanics, who are in Florida an overwhelmingly Republican constituency.[4.46, 4.47, 4.48]

• Representative Joseph Scarborough felt he had done exceptional work to place George Bush in office.[4.49] A Scarborough associate, Dennis Gilson, said that Scarborough might be rewarded with a cabinet post in the Bush Administration, suggesting that the work was exceptional indeed.[4.50] Scarborough's district included Escambia County in which voting machines in African American precincts showed astonishing rates of ballot spoilage of 10% and up - this in a county where the average rate of spoilage was 3.6%. 

• As stated, some counties obstructed our attempts to research what had happened.

  
  

C.  The electoral statistics are consistent with fraud. So far, the statistics are not consistent with alternative explanations:

• Through simple logic, excessive spoilage and strange values of crossover are exactly what is expected in the case of electoral fraud. While other explanations could also be valid, electoral fraud does serve to explain the observations and must be considered a reasonable hypothesis. 

• On average, on review of 3 million Florida ballots, there was approximately an 80% linkage between the party a voter chose for President and for US Senate in 2000, with Democrats being slightly more likely to vote a straight ticket.[4.51] So there is a fairly well-defined range of expected crossover ratios (defined as votes for Democratic president divided by votes for Democratic senator), between about 0.8 and 1.2. While there can certainly be regional variations due to local candidate popularity or unpopularity, this was not observed.

• The "Dixiecrat hypothesis," a key focus of disputes about whether peculiar statistics in northern counties can be explained by historical voting patterns, has been debunked in a paper titled Trends in Florida Voting. 

• The levels of statistical significance are very high, higher than one generally sees in the medical, physical, and biological sciences, to the point that one can call the pattern "systematic."

• As more counties were pooled into a larger meta-analysis, the level of significance continued to rise, suggesting that the two factors were identified were more important than other factors known to be involved in electoral outcomes, such as voting machine type and ballot design. 

So, statistical improbability is only a part of why we concluded that there was likely widespread electoral fraud. Based on what we know, a full recount would have meant that it would have been Al Gore’s election.  Presumably that is why the Republican Party had to "shut it down," in Congressman John Sweeney’s immortal words.

The big question is how did they know in advance that they had to "shut it down" in order to "win"?

5.   It would be impossible to keep a massive vote tampering conspiracy secret.

On conspiracies and recounts.  Critics are right to raise the question of how ballots from many different counties could have been tampered with without detection. At first glance, it may seem impossible that it could have been pulled off. However, in 2000, we questioned only about 7,100 ballots from 11 counties in a state where over six million ballots were cast and the outcome hung on just 527 ballots. Surely that would not require such a very large conspiracy.

But as the Iran-Contra scandal of the 1980s proved, even fairly large conspiracies can go undetected for many years if the watchdogs are lazy and the conspirators discreet. The claim that it would be impossible to tamper with, say, one ballot in a hundred without arousing suspicion does not hold water. Retail stores often write off that much as shrinkage. A few percent is certainly not "massive."  More to the point, it doesn't matter how many votes were stolen, if stolen they were. What matters is if there was knowledge of electoral tampering by one or both of the campaigns. No candidate who facilitates or silently permits self-government to be taken from the American people deserves to hold a position of leadership.

When one considers the detailed mechanisms by which electoral tampering can be done, it's obvious that there need not be any coordinated conspiracy. For example, suppose a county worker is charged with distributing ballots to a precinct with 70 Democratic voters and 30 Republican voters in an optical scan county. The worker decides to make small marks in the Bush oval on 20 ballots. If the voter who gets a pre-marked card votes for Bush, the vote will be counted. If the voter who gets a pre-marked card votes for Gore, the vote will be an overvote and will not be counted. So instead of the precinct turning out 70 votes for Gore and 30 for Bush, the precinct will have 56 votes for Gore, 30 for Bush and 14 spoiled ballots.

In the neighboring county, the technician who services the optical scan reader could align the machine such that it counts every Bush vote, but misses every twentieth Gore vote. And in yet another neighboring county in which tabulations are accumulated centrally without the precincts ever seeing the results, a county supervisor could alter the results at will, with no one being the wiser. With so much of the operation of vote counting in the hands of technicians from voting machine companies, it's possible for fraud to be committed without the knowledge of a single county resident. All it takes for electoral fraud to occur on a broad scale is for there to be strong partisanship, weak ethics, and lax oversight.

In 2000, there were recounts. There was, initially, a machine recount of many counties. Three large counties, Palm Beach, Miami, and Broward began full recounts, but only Broward was allowed to complete its recount. A statewide recount was ordered, but was interrupted by the Supreme Court's decision. We do not know what was discovered in that effort. 

The Miami Herald, The Orlando Sentinel, and National Opinion Research Center (NORC) undertook recounts of spoiled ballots. There are questions about the exact details of how each recount was accomplished.  Of the three, NORC's methodology was the most transparent. But as Douglas Jones noted in the report I critiqued, in some counties, there appeared to be a substantial mismatch between the number of spoiled ballots reported to the state and the number discovered on recount. And, of course, since there was no assurance that ballots were securely maintained, even a perfect match might not mean anything. A county supervisor, especially in some of the small, rural counties could presumably have removed or added ballots at will. All recounts, deficient as they were, showed a decided swing toward Gore.

Here is the key point: political and social leadership failed.   The situation in 2000 cried out for an investigation. Here, in a state which only 35 years earlier African Americans had been denied many of the basic rights of citizenship - denied many basic human rights - there was evidence that they and, as a consequence, tens of millions of other Americans had been denied the right to choose their political leadership. Countries in which people do not choose their leadership are called dictatorships, so the situation called for a serious investigation.

A few journalists and political leaders, notably Greg Palast of the London Guardian, Roger Roy and David Damron of the Orlando Sentinel, and Mary Frances Berry of the US Civil Rights Commission responded appropriately, using the resources they had available to try to find out what had happened.  A few academics, notably Allan Lichtman and Philip Klinkner, responded to the call. A few citizens, notably those who protested the inauguration, fulfilled their duties as citizens of a free country. But the vast majority of the country, especially its journalists, academics, and political leaders acted like the citizens of a country that was not free:  timid in criticism or vociferous in trying to silence critics.

The question of whether the election was fraudulent is almost of secondary concern. There are always bad people around, people trying to steal freedom, as though human rights were something that could be legislated away. But nations stay free as long as there is a "righteous remnant," willing to defend due process and the rule of law. Political leaders who could have acted and did not - candidate Al Gore, U.S. Attorney General Reno, Florida Attorney General Butterworth, journalists especially in the electronic media, academics in political science and statistics - these people could have forced a probing investigation that would have either dispelled or confirmed the fears of electoral tampering. They bear responsibility for what has ensued. But so does every citizen who stood on the sidelines, saying, "Defending the sanctity of elections is not my job."

The 2004 election.  In 2004, the reported election results were not nearly as close as they had been in 2000, yet the sense that something was amiss remained widespread. The exit polls, in particular, did not match the results. While more academics came forward, the response from journalism was again tepid.

Meg Laughlin and colleagues from The Miami Herald undertook to do a recount of three counties (Suwannee, Union, and Lafayette) and found what they thought was an acceptably close match to the original result.[5.1] This is certainly evidence against ballot tampering. But it is probably less conclusive than Laughlin seems to think. To start with, we were only able to obtain precinct data from Suwannee, so we don't know that the 2004 results from Union and Lafayette were suspicious. Also, we don't know what criteria Laughlin used to select counties for a recount. In particular, we don't know whether any supervisors declined a request for recount. It's possible that the three counties self-selected because they believed that a recount would show the count was accurate.

More important, Paul Lukasiak asked Laughlin if any ballots on which votes had not been recorded by machines had been hand-counted and, accurately or not, reproduced.  This practice was widespread in 2000, and although apparently legal, creates an obvious opportunity for significant fraud.  Inexplicably, Laughlin refused to reply. She also refused to supply me her data in electronic form, insisting that it could only be supplied by FAX and then only if I answered questions about why I wanted it. I declined her conditions.  

But even from the data in Laughlin's article, there appeared to be a small but unmistakable pro-Bush bias in the count. For example, she wrote: "The Herald total: 3,393 votes for Bush, 1,272 for Kerry. Fifteen votes couldn't be counted clearly. The county's official total: 3,396 for Bush, 1,251 for Kerry and a few dozen that couldn't be counted." In other words, the recount produced a 24 vote swing toward Kerry - that's half a percentage point! (Note also the vague language "a few dozen," which makes it impossible to get a firm grasp of discrepancies between the official count and the recount.) In Lafayette County, Laughlin's recount produced an 11 vote swing toward Kerry, just 0.1% of the total vote, but again in Bush's favor.

Laughlin's description of an apparently partial recount of Suwannee was so vague that it was impossible to be sure just what she did. Apparently she recounted the most heavily Democratic precincts. Apparently the results were close to what the county had reported. However, our work showed that unusual values of crossover, a possible sign of electoral fraud, were found primarily in Republican precincts. (In 2000, spoilage in Democratic precincts was a possible indicator of fraud, but spoilage in 2004 was far lower than in the preceding election).  So Laughlin may have chosen exactly the wrong precincts to recount.  

In fairness to Laughlin, she was looking to refute questions of vote tampering on a large scale that originated from the observation that Democratic registration was high and Democratic vote low in certain counties. She succeeded in knocking down that particular straw man. Still, this is the best evidence against the hypothesis of electoral fraud:  a partial recount of unspecified precincts for one county for which the investigator won't freely supply the data and will not answer questions about whether ballots were originals or reprints.  With such watchdogs, would it really be impossible for a few hundred Kerry votes to have been flipped to Bush in Suwannee County?

Laughlin was provided with a draft of this item from the FAQ and asked for her comments. She did not reply.

6.   Isn't statistics just BS? 

The public can be forgiven for thinking that statistics is mumbo jumbo. Battered with bogus numbers from The Tobacco Institute, the energy companies, and The Heritage Foundation, people have grown wary. That's good. But to people versed in statistics, there are only very rarely arguments about what the numbers indicate. The basic arguments revolve around the following:

• What level of statistical significance is meaningful? Although statisticians conventionally take 95% confidence as the benchmark, this is arbitrary.  One would probably want to have 99.9% confidence that one's food is safe. In preparing for a picnic, one might well accept 50% confidence that it won't rain.  Job seekers routinely accept 1% confidence or less in submitting resumes.

• Is variation really random?  One of the main assumptions of statistics is that variation follows the familiar Bell (normal) curve. While large excursions from normality are often permissible, and while methods can be developed for any distribution, standard methods can break down. To take a simple example, breast cancer is probably not normally distributed according to height. This is because breast cancer is rare in men, who also tend to be taller. So, in a graph of incidence of breast cancer according to height, the curve is likely to be skewed, with many more cases occurring among shorter people. If the results for men and women were graphed separately, however, one would expect both curves to be symmetric.

• What factors should be used to divide data into different groups?  This is often the most difficult question in any statistical analysis. One can always make an argument for saying that certain data are somehow "different," and ought to be treated separately from other data. One of the major issues debated on this website is whether the counties of north Florida are "different" from the rest of Florida. Certain counties do appear to be different in terms of the difference between how people register and how they vote. However, in crossover, there is no persuasive evidence that they are in any way unusual. If data are obtained from a controlled experiment, one can propose as many factors as desired, increasing the number of data points as needed. When the experiment is uncontrolled, post hoc assignment of factors can be deceptive. After all, if there's a finite chance of finding statistical significance by chance, if one invents enough factors, one is all but guaranteed of finding statistical significance where there is none. 

• What does it mean if two factors are found to be correlated?  If one found an association between height and breast cancer, it would likely be a false correlation. As pointed out, gender is related both to incidence of breast cancer and to height and is probably the primary factor in determining who gets breast cancer. Environmental exposure, genetic predisposition, diet, and other factors are probably also important. So, the simple fact that two factors are correlated does not mean that one factor causes the other. Small stature does not cause breast cancer. Still, even false correlations need to be understood and explained.

• When can statistical evidence be disregarded?  There are several cases when statistical methods are so badly flawed that one is justified in throwing up one's hands. First, post hoc methods that fail to consider what could be obtained by chance are justly disregarded as "data mining," if not outright obfuscation. Given enough factors, one factor will surely seem significant by pure chance. Second, if strongly-related factors are used for analysis, statistical significance can be lost. For example, if we include both gender and height in studying breast cancer, we may miss the association to gender, because a factor related to gender, namely height, may muddy the waters.  This issue is known generically as multicollinearity. Finally, false conclusions may be reached if distributions are very non-normal.

• Occasionally, statistical analysis is just plain wrong.  The usual benchmark of statistical significance is 95%. This means that one time in 20 the result is wrong by chance. For this reason, it pays to aim for much higher levels of significance. More important, one needs to crosscheck the implications of one's work. For example, if one concluded that height is the key factor in breast cancer, it would make sense to contrast incidence rates among groups known to be very short and very tall. It would also make sense to look at a primary mechanism underlying the proposed factor of height, such as pituitary function. False findings are often quickly unmasked by simple logic.  

Statistics is, above all, a tool to supplement logical analysis. Professional liars may be able to torture the numbers into giving the result they desire, but the flaws in their work become apparent by application of simple logical analysis. One of the silliest papers I have ever seen claimed that spoilage in African American districts is higher because African Americans sometimes refuse to vote for whites. As we have seen in Florida, spoilage in predominantly African American precincts - as in all precincts - plummeted with the introduction of new technology. Given this larger context, does the hypothesis of African Americans discriminating against white politicians really make sense as a primary factor in spoilage?

Did it ever? 

Here's the key question the reader should ask him/herself: "How certain do I want to be that an official holding office was really elected?"  Perhaps many people accept a large level of uncertainty. I would suggest that we should insist on very high levels of certainty, since the cost of uncertainty is political instability. The consent of the governed cannot said to have been obtained if tens of millions of Americans believe that an election was stolen, and the consequences play themselves out in many subtle, nationally-destructive ways.

7.   How did the work by Dawshed and Jordan avoid these pitfalls?

By dumb luck.

However, in reviewing how we approached the problem of analyzing Florida voting, I found that there were certain general features that explained how we happened to be so lucky. Most of all, we were very fortunate in having the work of Allan Lichtman and Phillip Klinkner on ballot spoilage as a good example. Although we (gently) tweaked Klinkner for the manner in which he introduced factors, and also suggested that his data may show a minor effect of education on spoilage, his work is beyond reproach.

So, here's why our work is sound:

• We established a clear logical link between strategies in electoral fraud and electoral outcomes, and defined our key variables (spoilage and pseudocrossover) appropriately.

• We looked at the role of many other factors, but at the county level. We eliminated factors as less important or because of multicollinearity (redundancy) at this point, and chose the two key variables before proceeding to end results. Only then did we begin the real analysis, which examined precinct-by-precinct data. Separating the design of a model from its testing by working at different demographic scales is an effective way of excluding personal bias from factor selection.

• We focused on as few factors as possible. As pointed out, with enough factors, one can end up with seeming-statistical significance purely by random chance.

• Our model worked better as more and more counties were brought into it, suggesting that it's a general model. When other factors, such as voting machine type were brought in, they did not improve the model.

• We went back and checked conclusions with past elections and with counties not included in the original analysis. We have considered how our results square with other observations such as the results of recounts.

• We made available our data and provided our analysis to other scholars; we actively solicited their opinions. Only two, Peter Caithamer and Walter Mebane, have raised a question or challenge specific enough to answer. We answered Mebane on whether certain northern counties have unusual crossover with a full-length paper, Trends in Florida voting. A brief report. Rev. 2.0. We seriously considered Caithamer's point on the normality of distributions, and will be addressing it in updates of A Model for Interpreting Voting Patterns with Application to FloridaRev. 1.0. 

No methodology, including our own, is immune from criticism. Real-life problems are infinitely variegated, and there are multiple methods to analyze them. These methods are typically very similar, but every statistical analyst has his or her own pet method, resulting in academic catfights of very little real interest. So, there's no rigid protocol to say how a problem should be approached, and that opens the door to those who maliciously use statistics against the public interest.

8.     So, what are the weaknesses of your work?

The greatest weakness arises from the huge amount of data, the fact that some counties report results in a convenient data format while some do not, and the fact that the data are scattered among 67 Florida counties, most of which have been helpful and a few of which have been outrageous and perverse. One would rather look at all of the data in one huge run. Unfortunately, given limited resources of (unpaid) time and computation, we did what we could. Offers of substantial funding to expand the effort will probably not be refused.

The issue of how closely spoilage, precinct by precinct, matches with Democratic vote vs. African American population needs a detailed examination. Is it more accurate to say that spoilage was targeted against African Americans or that it was targeted against Democrats?  One of the next projects is tackling Florida's notorious FREDS database to see if the relevant data can be extracted.

We also have not exhaustively examined the issue of whether the data follow a normal distribution. For small counties, it may not be possible to be certain. In 2004, spoilage almost certainly does not follow a normal distribution, so the preliminary analysis we performed in A Model for Interpreting Voting Patterns with Application to Florida Rev. 1.0 will need to be looked at again to see if other methodology would be more apt.

The variation of the crossover ratio by year also deserves close examination. There are certainly examples where it seems to vary widely between elections. Are there some counties in which it seems to be reasonably consistent?

We addressed the issue of turnout only superficially, not performing precinct level calculations. Recently I have been favorably impressed by work by Richard Hayes Phillips and Joe Knapp showing how to identify strategies of voter suppression through electoral statistics. While this work is not fully systematic, it is persuasive that voting machine shortages cost Kerry thousands of votes in Ohio. I believe it may be possible to generalize this work to include certain ballot destruction strategies.

Ellen Theisen and Warren Stewart (http://www.helpamericarecount.org/nm1.html) added an important element of analysis by examining the disparity between absentee votes recorded and signatures in the sign-in book, what they called "phantom votes." This is an important element of auditing that can detect fraud. On this site, Paul Lukasiak has also done excellent work, particularly in showing that the pattern of overvotes in Escambia County in 2000 was implausible.

Finally, and most important, there is the question of trying to match the statistical model against real-life observations. Repeated pleas for assistance from Floridians have not been answered. I get the sense that residents of north Florida are afraid of getting involved. Matching up the statistical model with newspaper accounts is similarly frustrating. In 2004, for example, Meg Laughlin of the Miami Herald was involved in a recount of three counties.[8.1] She declined to provide the precinct data to me by e-mail and refused to respond to an inquiry by Paul Lukasiak as to (a) what precincts had been recounted in Suwannee County, and (b) whether the recount had been done on original ballots or on copies, a practice that Lukasiak noted had only been revealed long after the 2000 election.  I will certainly report Laughlin's report on the recount as a fact at odds with the model Jordan and I developed. However, over the years I have seen so much deviation between what newspapers reported and what eventually emerged that I don't automatically assume her story of the recount is credible.

9.   Republican myths

Give the Republicans credit. They have been very persistent in spreading falsehoods to deflect attention and confuse people about what actually happened. Here's a short list:

10.    Non-Republican myths

The media and some academics have also been responsible for spreading certain myths. While not as obviously perverse and false as the Republican myths, they have served as blinders to prevent people from fairly evaluating what is before them. 

11.   You aren't objective. You have an agenda.

There are many in academia and journalism who think that being fair means giving equal time to NASA and the Flat Earth Society, to Galileo and to the Inquisition.

I am not one of these.

In defending what is plainly error, people who are faking objectivity are taking sides, indeed, actively supporting wrongdoing. When those people's paychecks depend on the approval of legislatures or press barons, it doesn't take much imagination to suppose that looking out for Number One might have influenced their critical thinking skills.  It's better to openly hold strong opinions. Better yet is to have opinions, yet be willing to change them in the face of facts and reason.   

I spent six months, a few hours a day, hunkered over a spreadsheet trying to make sense of Florida 2000 electoral statistics before I concluded there was evidence enough to believe that there had been fraud. I challenge anyone who says I am not objective to put in the same effort before making up his/her mind.   

If anyone were interested enough in my opinion to inquire, she/he might be surprised by the result. My opinion is that, according to facts were known at the time, Florida was a statistical tie: tiny errors inherent in counting millions of ballots made it likely that different outcomes might result from repeated recounts, even if the counters were scrupulously fair. Therefore, the logical outcome would have been to