PART II

WHAT IS

The Voting System

Voting is a system. It requires many steps: registering to vote, getting to the polls, casting a ballot, counting ballots, and certifying the vote. All the steps must come off without fail in order for a vote to count. All of the parts of the system must work well in order for the election result to reflect the will of the voters.
The challenge is to make voting less prone to error and more secure. In this section we consider the main components of the system. Subsequent sections detail the problems with specific elements of this system, beginning with equipment and registration, where we see the biggest problems lie.

An Overview

The voting system in the United States consists of four components: voter authentication, communication of voter preferences, the counting of these preferences, and security of the voting system.

• First, there is a method for authenticating voters: voter registration. Weeks or months before Election Day, eligible voters who wish to vote must register with the county or municipality in which they live. The local government compiles a list of registered voters and distributes that list (or at least the relevant parts) to the polling places. When voters come to vote, poll workers verify that they are indeed eligible to vote at their polling place.

• Second, there is a process for communicating preferences: balloting. To vote, people either go to public polling places on an appointed day and record their preferences on paper ballots or on voting machines, or people request an absentee ballot well before the appointed day. Americans vote using a wide range of different technologies, from paper ballots to touchscreen computers. Thousands of local governments and a few state governments make decisions about which voting technology to use and what the ballot will look like. A growing number of Americans (one in eight in the 2000 election) find Election Day inconvenient, and now vote "absentee" or "early."

• Third, there are procedures for counting ballots. For much of the nineteenth century Americans used paper ballots that were counted by hand; that system is still in use for about one percent of voters. Over the course of the twentieth century, voting equipment has evolved so as to speed up the count. These changes in technology have integrated the systems for casting ballots and counting ballots. Even with technology, however, many ballots are difficult to resolve. Because it can be difficult to determine a voter’s intention from the ballots and because machines fail, election laws in the states have evolved to clarify what counts and what does not.

• Fourth, there is a security system. To prevent coercion and vote buying, the states have adopted secret ballots. Local governments provide for the security of the count through public counting of the votes and inspection and auditing of the tallies by local canvassing boards. Electronic counting procedures (punch cards, scanners, and electronic voting machines) make the count difficult to observe.
  The replacement for the openness that paper ballots provide is a system of standards for electronic tabulation, developed and implemented by individual states or by the Federal Election Commission and implemented on a voluntary basis.

The four components of the voting system are supported by an extensive, decentralized administrative operation. Elections are conducted by the states. Almost all states have given the authority for administering the elections to local
governments. As a result, there are not fifty election divisions, but over three thousand election administrators maintaining voter registration systems, choosing equipment, formatting ballots, setting up polling places, handling absentee ballots, and conducting counts, audits, and recounts. The responsibility for paying for elections has also devolved to local governments. We estimate that all aspects of election administration cost counties roughly
$1 billion in 2000.

How Did We Get Here?

Why does the U.S. voting system have this particular structure? Much of the voting system today—secret
ballots, voter registration, machines instead of paper—evolved from reforms aimed at solving basic security
problems: the corruption of voters. Today, people make somewhat different demands; in particular, we ask that
it be more convenient.

Why can’t I have a receipt to check that my vote was counted?

This question cuts to the heart of the problems of how to design easy-to-use and secure voting systems.
A receipt is an easy check that every voter could use to make sure the process works correctly. However, receipts invite corruption. In the nineteenth century, we effectively did have receipts because ballots were not secret. The observable vote and, in some places, actual receipts allowed voters to trade their votes to local party officials and other political organizers for money, food, or alcohol.
Secrecy and anonymity of the ballot also provide important checks against coercion, against a person being forced, lured or intimidated into voting one way or another by others. In the late 1880s, almost all states adopted the secret ballot to combat widespread, organized vote buying. Receipts and other ways of violating secrecy raise the possibility of coercion.

Why don’t people vote?

Only about half of all Americans who are eligible to vote in fact do. There are many reasons why eligible voters do not vote. Many of these reasons have little to do with voting technology or the voting system at all. Some people are simply not interested in politics.
Many others say they are too busy or have difficulties getting to the polls. Even still, it is evident from studies conducted by the Census Bureau that many millions of registered voters who do not vote face obstacles to voting that could be lowered by correcting problems in the registration rolls or by making voting more convenient.
We make no promises about increasing participation.
Our concern is with those who show up and wish to express their heartfelt preferences, but cannot. Voters should not be excluded because the equipment did not work or because of errors in the registration rolls.

Why do I have to register in order to vote?

Voter registration is used to manage who votes in elections. Voter registration systems have been in existence for most of the history of the United States.
In the second half of the nineteenth century, voter registration systems became widely used to combat organized voter fraud in urban areas. Local political organizers coordinated “rovers”: people who would go from precinct to precinct and vote.

Roving voters highlight two problems that registration aims to solve. First, registration systems are intended not only to ensure that voting is confined to eligible participants, but also to ensure that voters vote where they are supposed to. Representation in the U.S. is based on geography: voters are allowed to vote for only those offices that cover their home. Each polling place is provided with a list of registered voters eligible to vote at that polling place. Second, registration is used to make sure that everyone votes once. If a voter can register only once, then he or she can only vote where the voter is registered; the election officer can, then, keep track of who has already voted and who has not.
My bank finds me no matter what.

Why can’t voter registration be as well informed? 

This is, in part, a consequence of decentralization.
Every county and state today has its own voter registration system, and voter registration is distinct from other county databases, such as motor vehicle registrations, drivers’ licenses, and taxation lists. So it is impossible for counties to keep track of voters. Several states have begun ambitious efforts at unifying registration statewide. This will ultimately produce cleaner voter registration rolls, by connecting registration to other databases, such as motor vehicle registrations and vital statistics. Such integration is very
expensive, but we believe that it will ultimately lead to a simpler registration and voter authentication system.

National voter identification cards are sometimes offered as an alternative to voter registration. Thanks to Napoleon, most European countries have citizen identification cards. These are used for voting, as well as many other government activities. The Anglo American countries—England, Canada, the U.S., and others—do not have such identification systems. Americans view national identity cards as undemocratic, giving the government too much ability to monitor us.

Why don’t we have a uniform method of voting like other countries?

Because of our system of federalism, elections are overseen in the U.S. by the states. The states have given local governments (mostly counties) the responsibility for day-to-day management of elections, while state governments check that the election was run properly, certify the official vote, and handle some administrative tasks, such as, in some places, registration.
Congress could impose uniform technologies for casting and counting votes in national elections.

Many other federal nations, like Canada, have separate national and local elections and separate methods for casting and counting ballots. For example, Canada uses hand-counted paper in the national elections, but some cities use electronics for their local elections. The U.S. tends to have uniform methods for casting ballots in each county for all offices. This allows us to have fewer days on which elections are held and to vote for more offices on election days. It has meant giving greater authority for election administration to the locales, and thus more discretion about voting equipment.

Do we really need technology to vote? Why don’t we just use paper and pencil like they do in Canada and France?

In the nineteenth century and well into the twentieth century most Americans did vote using hand-counted paper ballots. Most European countries still vote this way. Today only about one percent of Americans use hand-counted paper ballots.

Are Americans just fixated with technology?

The scale of U.S. elections requires technological solutions. In a European national election, where only the legislative election is on the ballot, there is just one vote to count. In a U.S. election, paper is very hard to manage, from the administrator’s perspective.
Paper ballots are expensive to print, secure, and transport. Counting is slow, labor intensive, and cumbersome, especially in many U.S. jurisdictions where there can be twenty offices and twenty ballot questions. The history of voting technology in the U.S., from handcounted paper to optical scanning and touchscreen computers, is the history of producing a speedier, more reliable count.

Why can’t I vote on the Internet?

Internet voting is here. The state of Arizona had one experiment with Internet voting in 2000, in the Democratic primary, and the Federal Voter Assistance Project ran a pilot project with the Defense Department for Internet absentee voting for overseas military personnel. We expect these experiments to grow, and the reason is simple: convenience.

Convenience voting is on the rise. Two decades ago only five percent of ballots were cast absentee or early; today that figure has grown to fourteen percent. The Internet is one of many technologies that can make voting more convenient.
However, Internet voting, in the judgment of many experts, is not ready for wide-scale use. There are three problems.

• First, there are concerns of coercion if Internet voting is done from remote locations, such as the voter’s home computer.
• Second, large-scale fraud is more likely because it is easier to hack the entire system if it is on the Internet, than it is to coordinate many millions of voters voting at precincts or thousands of poll workers.
• Third, many people do not have computers at home or are sufficiently intimidated by computers that Internet voting (either from home or at the precinct) might create a further obstacle to voting for millions of voters.
  Internet voting does hold immediate promise for lowering the obstacles experienced by some voters.

Technology today presents very significant obstacles to special classes of voters—most notably blind people (who cannot use visual systems and who have difficulty with transportation) and overseas military personnel (who cannot get to the polls and for whom traditional registration and absentee procedures are very difficult). The controversy over Internet voting and the answers to these other questions carry an important lesson. The way we vote is not static, and the decisions we make today will shape the future.

The voting system we have today evolved in response to specific problems. The most significant problems that have shaped our system were those of corruption and fraud, especially organized attempts to buy or steal votes. Fraud led to registration, secret ballots, and technologies for tabulation. Security considerations are fundamental to any changes made today in the voting system.

Today, there are additional problems, highlighted during the election controversy in the 2000 presidential election. We should have voting equipment that minimizes errors made by voters in casting ballots and that minimizes errors by machines in recording and counting ballots. We need a highly accurate and secure system for authenticating voters; currently, that is the voter registration system. We should have a very secure system for “convenience voting,” so as to guard against fraud in absentee ballots and to ensure that people who cannot be at the precincts can vote with confidence.

We should have a highly secure system for electronic transmission and tabulation of votes. We need a less ambiguous process for conducting recounts. And the U.S. has the opportunity now to lay the foundation for the future of voting.

Before we turn to what that future could be, we address the specific problems today, beginning with election equipment.

Equipment

RECOMMENDATION
Replace types of equipment that show high rates of
uncounted, unmarked, and spoiled ballots with
optically scanned paper ballots that are scanned at
the polling place by the voter (called “in-precinct
optical scanning”), or any electronic technology
proven in field tests.

Voting equipment was central to the election controversy in Florida in 2000. The recounts revealed many tangible problems voters had with ballots and machines and the resulting ambiguities in the tallies. Butterfly ballots and dangling chads instantly became part of the national lexicon.

But Florida was not unique. Florida had a relatively high rate of unmarked, uncounted, and spoiled ballots for president—three percent of all votes. Several other states, including Georgia, Idaho, Illinois, South Carolina, and Wyoming, had higher rates of unmarked, uncounted, and spoiled ballots. Some cities, including Chicago and New York, had rates of unmarked, uncounted, and spoiled ballots well in excess of the state of Florida.

The equipment used to cast and count ballots loses millions of votes nationwide each election. Over the past four presidential elections, two out of every one hundred ballots cast registered no presidential vote.
That rate is double in Senate and gubernatorial elections. Analysis of exit polls suggests that seventy percent of these uncounted votes are unintentional. In other words, approximately 1.5 million votes for president were “cast” but not recorded or counted in 2000.

Approximately 2.5 million votes for Senate and governor were “cast” but not recorded or counted over the last cycle.
The U.S. can cut the number of lost votes due to voting equipment in half by 2004 using equipment that is already available. We should replace types of equipment that show high rates of uncounted, unmarked, and spoiled ballots with optically scanned paper ballots that are scanned at the polling place by the voter (called “inprecinct optical scanning”).

As we document below, such in-precinct optical scanning has, on average, half the rate of uncounted ballots as punch cards and lever machines. In-precinct optically scanned ballots are not the only technology available today, but use of that technology could cut the rate of uncounted, unmarked, and spoiled ballots immediately.

In-precinct optical scanning is not ideal. It still loses votes. But it would represent a considerable improvement.
But we should also not lose sight of the future. Voting technology is evolving quickly. Many new machines are in development; they are untested but hold great promise. The best we can do today with upgrades is to reduce the average rate of lost votes in presidential races to about one percent of total ballots cast. We fully expect that new technology—technology that is currently in development—can reduce lost votes further and can break through other barriers in voting, such as handicapped accessibility.

A Provocative Scenario: It is 2002, and in a close U.S. Senate election, punch card ballots once again do not record a large number of votes unambiguously. The Secretary of State certifies a winner who holds a lead of 500 votes, among one million cast. The outcome of the race is in doubt. A recount is conducted, and a court battle over the count ensues.

What Equipment Do We Use Today?

Americans vote with five different technologies. These technologies differ according to the way votes are cast and counted.
Three technologies are based on paper ballots—handcounted paper ballots, punch cards, and optically scanned paper ballots. Hand-counted paper ballots are the oldest technology currently used in national elections. Nearly universal in the U.S. in the nineteenth century, they remain widely used today in rural areas.

Punch cards and scanners improve on hand-counted paper ballots by automating the count. Punch cards, which were introduced in the 1960s, require the voter to indicate his or her choice by making holes in a heavy stock card. Optically scanned paper ballots, which experienced explosive growth in the 1990s, require the voter to indicate his or her choice by filling in a circle or completing an arrow, much like answers to standardized tests are recorded.

These paper-based technologies differ in how they are counted. Election officials make tallies of hand-counted paper ballots. Scanning devices perform the tallies for the other two technologies. Card readers record the preferences of voters based on which holes appear in the punch card. Infrared optical scanners read the marks made on the scannable paper ballots.

Two other voting technologies involve machines that directly record the vote—mechanical lever machines and electronic voting machines (called Direct Recording Electronic machines, or DREs).

With a machine, the voter records his or her preferences on an “interface.” For the older lever machines, which were first introduced in the late nineteenth century, the interface is a set of levers associated with each candidate or answer to a ballot question. For the newer DREs the interface is a set of physical buttons or regions on a touchscreen that records a voter’s choices.
Whether the machine is mechanical or electronic, it unifies the casting, recording, and counting of votes in one apparatus. This has the advantage of eliminating the mass of paper that must be managed with paper-based systems.

Vendors often play up this particular feature of these systems, as managing paper is a big administrative headache for local election officials.
There are also important costs to the unification of equipment. Lever machines and DREs do not provide a separate record of the voter’s intent apart from that captured by the machines. Election officials can only recount what the machines record, so it is impossible to conduct a thorough audit of the election. And, probably most importantly, the user-interfaces are less familiar to voters than paper. This makes it especially challenging to design interfaces that do not confuse or intimidate voters. Because these machines are sold with their interface in place, only marginal improvements in the interface design can be made once the machines are acquired by local governments.

There are several important variations in the implementation of the designs of each of these five voting technologies. For instance, optical scanning is performed two ways—at the polling place (“in-precinct count”) and at the local election office (“central count”). In-precinct counts are widely thought to be superior because they give voters a chance to change their ballots to fix any mistakes detected by the scanner at the polling place.

Perhaps the biggest variations in design and implementation, though, are among the electronic machines. Older varieties of DREs are modeled explicitly on lever machines—they are essentially electronic lever machines. They present all choices at once (“full face”) on a large panel with push buttons. Such machines currently dominate the market, comprising approximately two-thirds of all counties using electronics.

Newer technology relies on touchscreens and keypads much like automatic teller machines at banks.
This technology is still infrequently used. It does have the potential to allow for upgraded and more flexible user interfaces (e.g., many languages).
Map 1 shows the great diversity of equipment used in the United States in 1999. Counties using hand-counted paper—the oldest system—are in white. This technology is used almost exclusively in rural areas today. Shades of red show counties that use the other paper systems—pink for punch cards and red for scanners. Shades of blue show counties that use machines—light blue for levers and dark blue for DREs. In some states, municipalities choose equipment and there is variation within the county. These states are shown in gray in the map. These data were collected by Election Data Services and by our project.

In the most recent election, only one in one hundred voters used hand-counted paper. One in three voters used punch cards. Slightly more than one in four voters used scanners. One in six voters used lever machines. And one in ten voters used electronic voting equipment. This pattern represents a significant change since 1980, when sixty percent of all votes were cast using lever machines or hand counted paper.

Over the past twenty years, local governments have increasingly abandoned traditional paper ballots and mechanical lever machines, in favor of methods that employ electronics in one way or the other, either to record the vote or count the vote or both. The Florida experience in 2000 has stimulated a number of states, including Florida itself, to abandon the first generation of computer-assisted voting, punch cards. There are, then, two types of technologies to choose between in the immediate future: optical scanning and electronics. How do they compare from the perspective of lost votes?

Voting equipment used in 1999

How Much Does Voting Equipment Contribute to Lost Votes?

Residual Votes and Lost Votes
Residual votes—the number of uncounted, unmarked, and spoiled ballots—provide a yardstick for measuring
the effect of different machine types on the incidence of lost votes.

Over the past four presidential elections, the rate of residual votes in presidential elections was slightly over two percent. This means that in a typical presidential election over two million voters did not have a presidential vote recorded for their ballots.

The presidential race is the “top of the ticket.” The rate of residual votes is even higher down the ballot—five percent for Senate and gubernatorial elections. In other words, almost five million votes are not recorded for other prominent statewide offices. A ballot may show no vote because the machine failed to record the voter’s preferences, because the voter made a mistake or was confused, or because the voter did not wish to vote for that office. The first two reasons would mean lost votes. The third would not be a lost vote, but would be a correct recording of the voter’s preferences.

It is difficult to judge intentions, but exit polls suggest approximately thirty percent of residual votes are intentional. This implies that 1.5 million presidential votes are lost each election; 3.5 million votes for governor and senator are lost each cycle.

A more conservative measure of the number of votes lost due to equipment is the number of ballots for which voters chose more than one candidate—an overvote. We focus on residual votes because the distinction of overvotes from other kinds of errors is a false one.

Technology can enable or interfere with voting in many ways. Lost votes are not just a matter of preventing someone from accidentally voting twice. Vote loss can happen because of machine failures. Vote loss also happens because ballot designs or user interfaces confuse voters or even obscure how to vote. Ballot and user interface design is perhaps the most important cause of vote loss, and different types of technology rely on specific types of ballots and user interfaces.

Whatever the cause, the residual vote rate should not depend on what equipment is used. But it does. 

The Relationship between Voting Equipment and Residual Votes

A simple table reveals the extent to which equipment affects the number of votes lost. Table 1 presents the residual votes in presidential elections and in Senate and gubernatorial elections as a percent of all ballots cast over the past decade.

The figures in Table 1 reveal a striking pattern. Some technologies consistently perform well on average, and some technologies have excessively high rates of residual votes. In particular, paper ballot systems tend to show lower residual votes than lever machines and electronic machines. To the extent that there is an exception to this pattern it arises with punch cards.
Optically scanned paper and hand-counted paper ballots have consistently shown the best average performance.

Scanners have the lowest rate of uncounted, unmarked, and spoiled ballots in presidential races and in Senate and gubernatorial races. Counties using optical scanning have averaged a residual vote rate of 1.5 percent in presidential elections and 3.5 percent in Senate and gubernatorial elections over the past twelve years. Hand-counted paper has shown similarly low residual vote rates.

Punch cards, the other paper based system, lose at least 50 percent more votes than optically scanned paper ballots. Punch cards have averaged a residual vote rate of 2.5 percent in presidential elections and 4.7 percent down the ballot.

Over thirty million voters used punch cards in the 2000 election. Had those voters used optical scanning there would have been 300,000 more votes recorded in the 2000 presidential election nationwide and 420,000 more votes in Senate and gubernatorial elections. Counties using paper ballot systems should choose either traditional hand counting or optical scanning in order to lower the number of lost votes.Machine voting, on the whole, has performed significantly worse than the paper systems.

Lever machines lost relatively few votes in the past four presidential elections, averaging a residual vote rate of 1.5 percent.
Electronic machines lost nearly as much as punch cards, averaging 2.3 percent over the past four elections.

The more severe problems appear down the ballot with these technologies, and here we see real concern with the continued use of lever machines. In recent Senate and gubernatorial elections, the average residual vote rates of lever machines and electronic machines were 7.6 percent and 5.9 percent, respectively, of all ballots cast. Had the counties using lever machines used optical scanning, we estimate that there would have been 830,000 more votes recorded in Senate and gubernatorial elections.

These patterns hold up to closer statistical scrutiny, holding constant turnout, income, racial composition of counties, age distributions of counties, literacy rates, the year of a shift in technology, the number of offices and candidates on the ballot, and other factors that operate in a county or in a particular year. For a fuller discussion see our report “Residual Votes Attributable to Technology: An Assessment of the Reliability of Existing Voting Equipment,” available at www.vote.caltech.edu 

The immediate implication of our analysis is that the U.S. can lower the number of lost votes in 2004 by replacing punch cards and lever machines with optical scanning. Punch cards and levers are, in our assessment, dominated technologies. That is, there are voting technologies available today that are superior, from the perspective of lost votes. Scanners consistently perform better than punch cards and levers. We also believe that optical scanning dominates older full-faced, push button DREs, which comprise fully two-thirds of the electronic machines in our analysis.

Touchscreens are, in our opinion, still unproven. Some counties, like Riverside, California, have had good experiences; other counties like Beaver County, Pennsylvania, and many counties in New Mexico had very high residual vote rates (over five percent in 2000).

This is not to say that optical scanning is an ideal system. It has plenty of faults and problems. This system also loses a significant number of ballots, though less on average than other systems. Election officials complain of paper jams, the cost of printing, and ballot management. Scanning is imperfect, but it is the best of what is.

For counties thinking of adopting optical scanning, there is a further question. Which sort of optical scan system is best? There are at least two different scannable ballots forms—connect the line and “bubble ballots.” Also, scanned ballots can be counted centrally (at the county election office) or they can be checked and counted at the precinct.

There is some evidence from the 2000 election, from states like Florida and Michigan, that precinct scanning has lower residual vote rates. Precinct scanning allows voters to fix their mistakes. The strengths and weaknesses of these specific aspects of scanning need to be more carefully and fully investigated before recommendations can be made.

We were most surprised by the comparatively poor performance of electronic voting machines. After all, we represent Institutes of Technology. One interpretation of our findings is that electronic voting is inherently flawed and should not be used. We disagree.

Electronic voting equipment has many apparent advantages. Unlike paper or punch cards it can be prohibited from registering overvotes. Unlike paper or cards, miscounting is virtually impossible. It is also possible to design interfaces for blind voters and to provide customized ballots on the spot.

We believe that the high rate of residual votes of DREs stems from the user interfaces. We have examined many of these machines. The mechanics of voting on these machines are often confusing. It is often not obvious how to undo a selection, how to check that all races have been voted, how to distinguish between the offices, and how to register the votes. Some interfaces are “too responsive”: a voter can push a button for the next page and more than one page will pass by without the voter seeing it. The formatting of the “ballot”—the presentation of choices—is often confusing as well. It is sometimes unclear where one office (a set of candidates to choose among) ends and the next one begins.

Ballot design is a problem with all equipment and lever machines, in particular.
We have also encountered physical reliability problems with some commonly used DREs, including lose connector cables that zero out the counters and blown fuses. Connector, cable, programming, and set-up problems can interfere with the conduct of elections. 

While the technology used is often excellent, the implementations have not always been at the level of other professional computer systems.
We see electronic voting as an improving technology.
It has great potential. However, in terms of one very basic requirement—minimizing the number of lost votes—electronic voting does not have a very good track record.

Paper systems have performed much better over the past dozen years. This problem means that the electronic voting industry is not working to the standards that it needs to. Our report holds this as a priority. It is unquestionably possible to make high quality, simple interfaces and manage complexity with computer technologies that exist today.

How Can Local Governments Acquire New, Expensive Equipment?

Election administrators must weigh not only the performance of equipment, but the cost of acquiring and operating their machines. The two viable technologies in the near term for most counties are optical scanning and electronic voting. What are the acquisition and operating costs associated with optical scanning and electronic voting?

Election Systems and Software, Inc. (ES&S), and Guardian Voting Systems, a division of Danaher Controls, two of the largest voting equipment vendors, provided us with information on acquisition prices and operating costs for different kinds of equipment. 
Their figures square with each other and with recent equipment purchases.

Assume that the life of these machines is fifteen years. The total cost of the equipment is the acquisition cost plus fifteen times the operating cost. The total cost of a touchscreen DRE system comes to approximately $32.75 per voter over the entire fifteen year span. (We use $21.50 for the acquisition cost.) The total cost of an optical scanning system comes to $29.50. (We use $7.00/voter for the acquisition cost.) Even though optical scanning systems have much higher operating cost, the difference in the acquisition cost is sufficiently large that the total cost of the optical scanning system is somewhat lower over the fifteen-year operating life of the machinery. If we assume a twenty year lifespan, the costs are identical.

For an election administrator these numbers seem daunting. A city with 250,000 registered voters would spend $5 million to purchase equipment. This sum exceeds the total election administration budget of a city this size. Leasing is one possible solution, as we discuss later in “Cost and Public Finance of Elections.”

Other Considerations 

Reducing the number of lost votes is a very important goal, but it is not the only factor in choice of equipment. Security and misvotes are also important, though we know of no data on these factors. Three further considerations are auditability, management, and accessibility. 

Auditability

In the 2000 presidential election, the state of Florida conducted an enormous audit of its voting machines. It checked the record of the vote cast—the punch cards and scanned ballots—against the final tally. It is extremely important to be able to conduct such an audit. So long as we can verify the official count through a systematic recount of the votes we can avoid having to call an entirely new election, a revote.

Paper ballots have the highest degree of auditability.
The voter records on paper what he or she intended.
This can always be examined in a recount, if it has not been lost or stolen.
Lever machines and older direct recording electronic machines offer no auditability. If a machine is jammed or broken, the recorded tally will not reflect the votes that people cast. The votes cast on a broken machine can never be reclaimed. For this reason alone we feel that lever machines and older DREs should not be used.

Most new electronic machines produce an internal paper tape (like a cashiers tape) and an electronic recording of every voting session. This allows officials to reconstruct what was done on the machine. While this is an improvement over other machines, it is not a direct recording of the voter’s intention. If the machine fails between the touchscreen and the tape, the voter’s stated intentions are still lost.

We feel that new voting standards must require a minimum level of auditability. The industry is searching for such a standard on its own, mainly through demand from local election administrators. This is a situation, though, where clear standards should be set nationally; the equipment industry can build to those standards.

Management

Managing ballots and equipment on Election Day is a Herculean task. Little things can happen that are difficult to control but that produce lost votes. One of the more alarming stories in Florida involved a poll worker who accidentally took home a bag of ballots, thinking the bag was his laundry. There was no malicious intent, but the example shows how insecure the ballots really are and how difficult it is to keep track of all ballots on Election Day.

Different technologies pose different management challenges. Machines, especially lever machines, are costly to store, maintain, and deploy. Paper ballots —hand counted, scanned, or punched—must be transported and processed, an especially difficult task if ballots are counted centrally. Los Angeles County, California processes 2.8 million ballots in one night.
County election officials must coordinate the transportation and counting of all those ballots.

Accessibility

One of the most challenging problems facing voting today is making voting accessible to all eligible voters. Today there are two obvious and difficult obstacles: disabilities and language. People with disabilities often cannot vote without assistance. There are two million blind people in the United States, none of whom can vote without assistance. People who do not speak English with comfort or who are illiterate often cannot vote without assistance.

The voting equipment industry has been grappling with these problems in recent years. It has made some progress developing machines that are usable by blind voters. Many new DREs offer recorded instructions on how to vote. The voter must still navigate the touchscreen or push button. This represents a very important advance, but we know of no studies of the performance of these machines. We strongly recommend human testing of equipment for errors in voting and ease-of-use of equipment accessible to blind voters.

New interface designs and machine architectures may be needed to solve accessibility for blind voters and for voters who need assistance reading English. We think the best approach to addressing these problems involves federal investment in research and development of appropriate designs and equipment.

Registration

RECOMMENDATION

Near Term

• Develop a system for allowing voters to check their registrations.
• Develop better databases (e.g., record some sort of numerical identification on each voter’s registration).
• Make the county’s or state’s registration database accessible at each polling place.
• Provide polling places with the list of dropped voters and the reason they were dropped.
• Use “provisional ballots” aggressively when there are registration problems. 

Long Term

• Computerize voter registration information and processes at both the local and state levels.
• Develop statewide qualified voter files.
• Fix gaps in the more open registration system created by NVRA.

Along with the secret ballot, voter registration provides a basic check on the integrity of voting in the U.S. Registration does two things. First, registration information is used to control control who votes. Only those who are eligible to vote can register. Poll workers use the registration rolls to authenticate voters at the polling places. This is a check on roving voters, non-citizen voting, and other abuses. Second, registration information is used to manage ballots.
The addresses on the registration lists determine where people are eligible to vote and, therefore, which ballot a voter is supposed to receive.

Voter registration is essentially a state census, administered locally, and developed exclusively to manage voting. Performing this census is a daunting task. Start with the numbers. The number of potentially eligible voters in the United States—the “voting age population” (VAP)—is over 200 million. The voting age population grows about two percent nationwide every two years. In the four years between presidential elections, local election officials have to deal potentially with four million new voters simply due to the natural increase in the population.

National legislative changes in the 1990s added significantly to the burden local election officials face in registering new voters who come of age, removing those who die, and handling changes of address. Most significant was the National Voter Registration Act of 1993 (NVRA), or “Motor Voter,” which imposed many new requirements on local officials, in an effort to make registration itself more convenient and to make it more difficult to purge inactive voters from the rolls.

The added convenience of registration has encouraged the number of registrations to grow, to the point where the number of new registrants is vastly outstripping the natural increase in the number of eligible voters.

For instance, between 1994 and 1998, the size of the eligible voting population population grew by 4.3 percent (8.3 million people); over that same time the number of registrants grew by 19.6 percent (25.7 million people).

However, this big increase in registrants has not produced a concomitant increase in the number of voters. Consequentially, one immediate effect of the NRVA has been to increase the number of “inactive” registrants, from 1.7 million in 1994 to 14.6 million in 1998. Registration is a significant, never-ending task. With the promise of expanded voter participation, the NVRA also brought new administrative headaches that are only now beginning to be adequately addressed by states and localities.
To manage this task, a voter registration system must meet five standards.

• First, registration information must be accurate and complete. The information on the voter registration rolls must cover all registered voters and have the correct information used to authenticate the voters, that is, to verify that the voter is eligible to vote for a prescribed
  set of races.
• Second, registration information must be immune from fraud. If the aim is to prevent fraud, then it should be difficult or impossible to create fraudulent registrations.
• Third, registration information must be dynamic and up-to-date. Voter registration must be flexible to accommodate frequent moves made by previous voters, the addition of new voters, and late voter registrations. Registration must also fit with election schedules. A significant challenge is developing a fraud-resistant system for last-minute registrations, including Election Day registration.
• Fourth, registration information must be usable by the election officials at the polling places. Because election officials use this information to authenticate voters, polling place workers must have usable registration information.
• Fifth, it must be easy for voters to register. Registration should not be a burden to voters.

When we began this project, election administrators told us that their biggest problems lie in the area of registration. Problems maintaining the registration system make it very difficult to control who votes and to manage ballots on Election Day. These are very big problems.

• First, errors in the registration rolls prevent some people from voting. Registration is a large database management problem. As in any database, errors can occur many ways. Voter registration databases suffer from typographical errors, dropped registrants, and outdated information. Nationwide, the Census Bureau estimates that in the 2000 election three million registered voters did not vote because of problems with their registrations.
• Second, fraudulent or outdated registration may allow fraudulent voting. People who are not eligible to vote may try to register. Examples of such fraudulent registrations include registration by non-citizens or registering multiple times. People may use other people’s registration information to vote. The most notorious examples involve recorded votes by dead people. It is unknown how much fraudulent voting occurs because of registration failures. One study, sponsored by the Atlanta Journal and Constitution, discovered that fifteen thousand dead people were on Georgia’s voter rolls, out of a total of 3.6 million registered voters. Over a twenty-year period, 5,400 dead people were discovered to have voted in Georgia.

Audits of voter registration systems have found astounding numbers of duplicate registrations. Los Angeles County, California recently audited its registration rolls and found that one in four registrations were duplicates (usually because people moved).
When Michigan updated its voter files, the state discovered one million duplicate registrations (out of nine million registered voters). There is little evidence that such duplicate registrations have led to widespread duplicate voting.
Improvements in the accuracy of registration systems are needed in order to prevent denial of access to the polls, to prevent significant fraud, and to assure legitimate voters that their votes are not diluted.

A Provocative Scenario: It is 2002, and in a close U.S. House election, 5,000 potential voters, out of 100,000 cast, claim they were turned away from the polls on Election Day. Almost all of these were people who registered to vote when they renewed their driver’s license. Further investigation reveals that the local election supervisors had not processed a backlog of registration forms that had arrived well before Election Day.
Enough qualified voters were turned away that the courts declare the initial election result invalid. A revote is ordered. The declared winner of the original election challenges the revote.
The House of Representatives, in which the majority party holds a two-seat advantage, must decide whether to seat the original winner, go along with the court-ordered revote, or follow some other course to settle the election dispute.

Why Do These Problems Exist?

• Perhaps the most important explanation is that registration is a massive, complex database. In any system large enough to keep track of 150 million registered voters there will be typographical and other data errors. Changes in the population complicate matters. Americans move a lot. In March 2000 the Census Bureau estimated that over fifteen percent of eligible voters had moved in the previous year. So the rolls are in constant flux. 
• A second factor contributing to the problems with voter registration is that it is decentralized. Management of the voter registration system is handled in most states by local governments that do a good job with limited resources. Duplicate registrations and other problems, however, emerge because there is no method for coordinating these local governments’ registration databases. If a voter moves, there is no ready method in most states for updating that voter’s registration, apart from the voter taking the initiative to change registration in both counties. This is a hassle. Even within a county, people move without updating their registration. Some of these people try to vote at their new addresses but cannot. 
• A third factor is that voter registration information is difficult to deploy on Election Day because of the precinct voting system. There are well over three thousand jurisdictions that manage voter registrations, but there are approximately 200,000 polling places, each of which needs access to the registration information.
  Almost all counties distribute registration information to the polling places by printing out the list of people who are registered and eligible to vote at a specific polling place. Handling problems at the polling places is very time consuming: the poll worker typically must call the central election office to verify registration information whenever there is a problem. This
  distracts from other activities at the polling place, including attending to voters who need assistance.

Addressing these problems is a continuing activity of localities, and increasingly the state and federal governments. The most significant recent federal legislation is the NVRA. This law set standard procedures for purging registration rolls and allowed voters to apply to register at departments of motor vehicles and other public offices.

The NVRA lowered many barriers to registration and addressed many civil rights problems. But, it may have exacerbated database management problems. Many registration applications do not make it to the local election office. As a result some people think they are registered where they are not.

The U.S. must continue its efforts to improve registration. We have the following concrete recommendations toward this end.

• First, develop a system for allowing voters to check their registrations. This might be done by publishing all registrations in a local newspaper at least a week in advance of the closing of registration or by sending post cards to all registered voters or to all residences with the current information. Some counties in North Carolina now allow voters to verify whether they are registered to vote via the Internet.
• Second, develop better databases. A simple step is to record some sort of numerical identification on each voter’s registration. We recommend driver’s license numbers or the last four digits of the Social Security number. Only fourteen states require information that could be used as a database index, though over half request such information. Such an index is essential for managing purges and duplicates. With such an index the state could verify whether registrations with common names (like Joe Smith) are duplicates.
• Third, make the county’s or state’s registration database accessible at each polling place. We recommend putting the complete registration database for a county on a compact disk and leasing a laptop computer for each polling place. Where this has been done it has reportedly eliminated a majority of registration problems and reduced polling place bottlenecks.
• Fourth, provide polling places with the list of dropped voters and the reason they were dropped. Many registration problems arise because of incorrectly purged rolls. Even without providing the countywide information, these problems could be fixed by providing the list of purged voters.

Finally, counties should use “provisional ballots” aggressively when there are registration problems. A provisional ballot is a “fail safe” method that can be used when a potential voter’s registration status is challenged at the precinct. A voter who votes a provisional ballot is allowed to make choices among offices that are common to all voters in a county, including all statewide and county offices, and possibly state legislative offices, too. The ballot is then sealed in an envelope, along with an affidavit from the voter declaring that he or she is eligible to vote. After Election Day, the registration status of the voter is verified. Individuals who should have been allowed to vote then have their ballot counted. Individuals whose registration does not check out have their ballots discarded.
We estimate that aggressive use of provisional ballots could itself cut the rate of lost votes associated with registration problems in half. Currently two-thirds of the states do not use provisional ballots, and many locales that provide for them do not use them aggressively.
In Los Angeles County, California, two-thirds of the provisional ballots that were issued on Election Day in 2000 were valid ballots. These two facts suggest that aggressive use of provisional ballots could cut lost votes due to registration problems by half nationwide. That is roughly 1.5 million lost votes.
We must also consider long-term changes in the registration system. Stop-gap and fallback measures, like provisional ballots can alleviate problems, but they represent superficial corrections for deeper problems.

• First, the counties and states should computerize voter registration information and processes at both the local and state levels. Many states, like Michigan and Oklahoma, have already begun such a process. It is essential for guaranteeing the integrity of the voter registration rolls. It is a very expensive process. Federal funding could help in this process.
• Second, develop statewide qualified voter files. Several states have begun to develop such files. This would allow for thorough checking of duplicates, and may make it easier to detect fraud.
• Third, we must fix gaps in the more open registration system created by NVRA. Some states and locales have integrated their voter registration databases with other county and statewide databases, especially those agencies relevant under the NVRA. We are concerned about the procedures for third-party registrations. Some organizations that solicit new voter registrations never forward the registration forms to election registrars. Some organizations may even use the prospect of registration as a way of collecting information about people. One might allow only official government offices to conduct on-line voter registration.
• Fourth, develop electronic authentication of voter registration at polling places. We estimate that leasing a laptop for Election Day costs $100, and the wages of a county employee in charge of the laptop would cost $400. With two elections per year the cost comes to roughly $2 per voter per year. 
  States and counties have already initiated reforms along these lines. Several examples of reforms and best practices deserve to be highlighted, especially as other states and locales can learn from these experiences.
  Included among these are the Michigan Qualified Voter File, the California “on-line” voter registration process, a program in Orange County, Florida to use laptop computers to deal with registration issues in the precincts, and the Federal Voting Assistant Program’s 2000 “Voting Over the Internet” pilot program.

NEXT PAGE  CALTECH INDEX

Farming Politics Government Posters Humour Technology Religion Nature Me Links