CalTech 2

On December 15, 2000, the California Institute of Technology and the Massachusetts Institute of Technology announced a collaborative project to develop new voting technology in order "to prevent a recurrence of the problems that threatened the 2000 presidential election." The problems in the 2000 election go well beyond voting equipment. This report assesses the magnitude of the problems, their root causes, and how technology can reduce them. We call for a new architecture for voting technology that is tailored to the communication and computing technologies that have revolutionized our society. We also see a new system of continual innovation that can be supported by the federal government.

Our data show that between 4 and 6 million votes were lost in the 2000 election. Our analysis of the reliability of existing voting technologies and election systems shows that the U.S. can substantially reduce the number of lost votes by immediately taking the following steps:

• Upgrade voting technologies. Replace punch cards and lever machines with optical scanners. We estimate 1.5 million of these lost votes can be recovered with this step.

• Improve voter registration systems. We recommend improved database management, installing technological links to registration databases from polling places, and use of provisional ballots. We estimate this could save another 3 million lost votes. Aggressive use of provisional ballots alone might substantially reduce the number of votes lost due to registration problems.

In the long term, the voting equipment industry will develop new technologies. Our report includes the following recommendations to ensure that the best available technologies are developed by this industry:

• We call for a new architecture for voting technology. This architecture will allow for greater security of electronic voting. It will allow for rapid improvement and deployment of user interfaces—that is, better ballots. It is a framework within which we can explode several myths about electronic voting.

• There must be significant investment by the federal government in research and development of voting equipment technologies and meaningful human testing of machines.

• The federal government should establish an independent agency to oversee testing and to collect and distribute information on the performance and cost of equipment.

The furor over the 2000 presidential election in Florida brought this group together. David Baltimore, the president of the California Institute of Technology, and Charles Vest, the president of the Massachusetts Institute of Technology, assembled our team of computer scientists, mechanical engineers, and social scientists to consider what is and what could be. The Carnegie Corporation sponsored our project.

This report offers our assessment of what works, what does not, and what can be improved in existing voting technology. How big are the problems in voting? What solutions exist today? How can we improve voting for the 2004 presidential election?

Our ultimate goal is to develop ideas about what could be. The United States is in the midst of a revolution in communication and computing technology. That revolution will transform voting in the future. These technologies hold enormous promise—to make voting easy, convenient, and accessible, and to allow voters to see that their votes are counted.

In addition to the faculty involved, many students contributed to this project. Darian Unger, Jonathan Goler, and Aaron Strauss provided invaluable assessments of user interface designs.

Tara Butterfield, Lee Carpenter, Michelle Nyein, Meena Untawale, James Wagner, and Catherine Wilson helped collect data on public finances, election results, and machine usage in the United States.

Our group was assembled for its expertise, rather than its political leanings. Some of us are Democrats; some of us Republicans; some of us have no partisan leanings or political inclinations.

Two professional staff members have coordinated our activities and made this project happen. We owe a special debt to Julie Brogan, Esq. and Mary King Sikora.

The controversy in Florida exposed two very important problems with the way elections are run
in the United States: recounts and system failure.

Contested elections happen. In the event of a contested election, candidates can challenge the
initial count and request a recount. If there are sufficient problems, especially stolen or fraudulent
ballots, the courts may have to resolve the counts or even require a revote.

The Florida recounts demonstrated just how hard it is to determine who won—given the existing means of casting and counting ballots. Some technologies produce particularly poor records of the voters’ intentions. The controversy in Florida centered on punch cards. Many votes were lost because voters did not punch the card through entirely or they punched two candidates’ names, perhaps by accident. These problems were widely blamed on the voters, though similar voters had fewer problems with other technologies, such as in-precinct optical scanners.

As the challenge in Florida moved first to the election boards and then into the courts, it became
evident how very difficult it would be to resolve the count. The technological flaws could not be resolved unambiguously by recounting using the ballot counting machinery. Election officials and judges had to make judgments about what should be counted and how, and then they had to count the ballots anew, by hand. Lacking clear legal standards, votes were not considered and counted the same in different jurisdictions.

Machinery that loses votes is worse than machinery that produces ambiguous records of voter intention.
Lever machines and many electronic voting machines provide no record of voters’ intentions apart from the count itself. If a machine is broken, say because of a jammed counter or an electrical short, then all votes are lost on that machine. If a sufficient number of votes are lost, the election is thrown into question. The election may also be thrown into doubt if an act of fraud alters or destroys a sufficient number of votes. Then, we may have to conduct the election again.

How common are these problems? How many elections might be affected? How many jurisdictions might be affected?

Close elections, problematic votes, and recounts occur in every election year and in every jurisdiction. In 2000, Florida’s presidential votes were recounted, so, too, were U.S. Senate votes in the state of Washington. In the 2000 presidential election, the winner’s margin was less that one-half of onepercent in four states: Florida, Iowa, New Mexico, and Wisconsin. Many state legislative contests were recounted as well—for example, three in Colorado.

Recounts occur when the margin between the top two vote getters is extremely small. We find that the rate of uncounted and spoiled ballots ranges from two to three percent in presidential contests, depending on the equipment used, and from three to seven percent in Senate and gubernatorial elections, depending, again, on the equipment used. This rate of questionable
ballots is high enough to affect numerous contests each year.

The problems observed with specific technologies in the 2000 elections are not new. Following widely publicized problems with punch cards in the 1968 election, IBM withdrew from the election machine business. In 1996, a contested and recounted primary for the Massachusetts 10th Congressional District led that state to abandon punch cards. We now face a similar choice nationwide because of the problems with punch cards in the Florida recounts. Technologies
available today can produce better records of the vote than punch cards. These other technologies, while superior, are, nonetheless, imperfect. In order to minimize ambiguous recounts, we need to improve voting equipment.
Contested elections happen. We should prepare for them.

System Failure
Recounts and contested elections aside, it is important to make sure elections work. They are the foundation of our democracy and a model for democracies around the world.
Like most Americans, we took it for granted that election administration worked. Surely, after so many years of successful democracy, the nation had settled on a reliable means of casting and counting votes.

The scrutiny given to the vote in Florida opened our eyes to the very real possibility that, in the United States today, many votes are not counted. Many registered voters evidently went to the polls, cast ballots, and those ballots, for whatever reason, could not be counted. Still others made every attempt to vote but could not.

Journalists’ investigations around the United States revealed that Florida was not the worst state and that Palm Beach was not the worst county. Illinois, South Carolina, and Georgia all had higher rates of spoiled or uncounted ballots in the 2000 presidential election. In Chicago, almost one out of every ten ballots for president did not register a vote. Chicago’s problems were not limited to punch cards. From New York City came reports of improperly printed ballots and broken lever machines. From Beaver County, Pennsylvania, came reports of high numbers of unrecorded ballots using a new touchscreen computer voting system. From New Mexico came reports of voting disrupted by bad weather and power outages.

We are concerned about the potential long-term effects of such problems on Americans’ confidence in their own electoral process. To many Americans this string of stories was just more bad news about the workings of American government, more reason not to vote, or more reason to turn away from public life. We cannot measure the strength of voters’ confidence in the system. And, in fact, we think it would be foolhardy to wait for public opinion to sour before addressing the problems in the voting system.

Once we lose confidence in a system as fundamental as voting, it is too late.
Today, many parts of the government are working to improve the voting system, with the aim of restoring voter confidence. In response to the controversy in Florida, many states and the federal government have initiated significant efforts to reform election administration.
Local election administrators work continually to make their voting systems better. They are people personally dedicated to making democracy work.
Where to begin? How big are the problems with voting equipment in the United States? How do these problems compare to the other aspects of voting, such as registration? And what will it take to fix the components of the system and the system as a whole?

Lost Votes
We estimate that between four and six million presidential votes were lost in the 2000 election These are qualified voters who wanted to vote but could not or were not counted. Losses occur for two reasons: first, some voters do not, or cannot, participate due to problems with voter registration or polling place practices; second, some votes that are cast are not counted due to problems with ballots.

Two million ballots, two percent of the 100 million ballots cast for president in 2000, were not counted because they were unmarked, spoiled, or ambiguous.
Of this two percent it is estimated that 0.5 percent did not intend to vote for president, so 1.5 percent (or 1.5 million people) thought they voted for president but their votes were not counted. Below the office of president the incidence of spoiled, unmarked, and uncounted ballots is much higher: five percent of ballots do not record a Senate or gubernatorial vote. And there are significant differences across equipment types in the incidence of uncounted ballots. For example, since 1988, three percent of voters using hand-counted paper and scanned paper ballots had no vote recorded for Senate or governor, but seven percent of voters using lever machines recorded no vote for Senate or governor. Thinking only about elections for Senate and governor, the differences across equipment add at least one million more votes to the numbers lost.

We lost between one-and-a-half and three million votes because of the registration process in 2000. According to the U.S. Census, Current Population Survey, 7.4 percent of the forty million registered voters who did not vote stated that they did not vote because of registration problems. Voter registration is an enormous database management system—a local census, if you like. Errors in databases occur even under the most scrupulous management. In practice, voters
are not always careful in filling out registration information or in keeping their registration information current. We lost between 500,000 and 1.2 million votes because of polling place operations. According to the U.S. Census, Current Population Survey, 2.8 percent of the forty million registered voters who did not vote in 2000 stated that they did not vote because of problems with polling place operations such as lines, hours, or locations. The figure was 1.2 percent in 1996.
We lost an unknown number of votes because of mishandled and controversial absentee and overseas military ballots.

4 TO 6 MILLION LOST VOTES

1.5 to 2 Million Lost Because of Faulty Equipment and Confusing Ballots

1.5 to 3 Million Lost Because of Registration Mix Ups

Up to 1 Million Lost Because of Polling Place Operations

Unknown Losses Because of Absentee Ballot Problems

The equipment figures come from our own analysis of lost votes. The registration and lines figures come from a survey conducted by the U.S. Census. We are more confident in the equipment figures. We take the survey figures at face value. They may be too high, owing to ambiguous wording in the Census question.
We consider the survey data in the Appendix.

The recounts in Florida revealed another sort of error: mistaken votes and incorrect counts. The butterfly ballot confused many voters, producing mistaken votes. Thinking they voted for Candidate A, many people accidentally voted for Candidate B, because of the confusing layout of the ballot. Accounts of the recount noted that each time the punch cards were run through the counters, the tallies differed.
A systematic analysis of the rate of errors made by voters and by tabulation machines is needed to measure how large these problems are. This will require extensive testing and experimentation with equipment and ballots. That should be part of a larger, federal program on election administration.
In the area of tabulation, existing voluntary standards represent an important improvement. Existing standards developed in 1990 set minimum criteria for tabulation errors (on the order of one in 250,000 or fewer) of any new vote counting equipment. These standards are voluntary, and testing is performed on machine prepared ballots, rather than on ballots that are marked by people.

Security of ballots and counts is a different sort of problem than lost votes or incorrect votes.
Fraud is, by its nature, hard to detect and measure.
Stealing votes is a targeted attack on the electoral process, and those attempting fraud try to cover their tracks. This makes reliable measures of fraud tricky.

Fraud and security are social problems—people will commit fraud if they are willing to win by any means.
Error is more of an engineering problem; we should make every effort to make machines, databases, and
other aspects of the voting system more reliable. The social nature of security also means there are different solutions available. Penalties for electoral fraud and improved detection methods can act to deter individuals from conducting fraud. Judging by recent court cases, the greatest fraud problems may lie in absentee balloting (though registration also presents some problems), a part of the process that has less oversight than voting in precincts.
One cautionary note: some technologies are on their face suspect from the perspective of security.
We are particularly concerned about the prospects of disruptions of voting over the Internet. A single attack
targeted against Internet voting could have much bigger consequences than the diffuse sort of activities required to defraud precinct-based voting. We are also concerned about the secretive and proprietary treatment of tabulation software of all electronic voting
The fraud that occurs one ballot at a time or one lever pull at a time accumulates slowly like grains of sand on a scale. We are more concerned about someone putting his or her thumb on the scale.