Polling Places

RECOMMENDATION
Election Administrators should measure the performance of individual polling places in the areas of arrival process, authorization to vote, voter education, and staffing practices and adopt management principles to improve service.

Polling place lines are part and parcel of every Election Day news account, and 2000 was no different. A legal tug-of-war happened in St. Louis in the 2000 election, and it may have affected the outcome of the U.S. Senate election in Missouri. What we find especially troubling about polling place lines and closings is that voters who have done everything right are denied access to the vote. Voters who register, study the choices, make the effort to go to the polls, and arrive on time can be denied the vote because of unusually long lines at the very end of the day.
According to the U.S. Census, in 2000, 2.8 percent of registered voters who did not vote said that they did not vote because the line was too long or the hours were too short. That is approximately one million voters.

Polling place set up is a logistical challenge.
The typical polling place handles 400 to 500 voters on Election Day. There are approximately 200,000 polling places in the U.S. on Election Day, staffed by 700,000 employees hired just for the day. The pay is minimal.
The polling place is a service system; it provides the service of voting. The voter is the customer with certain requirements. Namely, the voter wants to cast his or her vote accurately, privately, with minimal wait, and with absolutely no hassles. The mission of the polling place should be to satisfy its customers, spending the minimal amount of resources needed to do so.

We believe that polling place service can be made better, possibly lowering lines, by reorganizing staffing.
The polling place can be viewed as a queuing system, comprised of a series of queues. Voters arrive at the polling place and first enter a queue at which they get authorized to vote, by means of a check on their registration. If there were a paper ballot, the voter typically would receive it at this point. Also, there might be an opportunity for education on the mechanics of filling out the ballot or of operating the voting equipment.
The voter then enters a second queue to wait for a voting booth to vote. The voter then goes to a final queue at which he or she deposits the ballot (e.g., an optical scan machine) and checks out.

Three important characteristics of this system are as follows.

• First, the arrival rate to the polling place varies dramatically over the course of the day. There are three major peaks—early in the morning, the lunch hour, and early in the evening after work.
• Second, poll workers are primarily volunteers, who work the entire day, from when the polls open (typically 7:00 a.m.) to when the polls close and the votes are tabulated (typically 8:00 p.m.–9:00 p.m.).
  The majority of poll workers are elderly persons who have retired from the work force.
• Third, voting is not a frequent occurrence, and voters have limited experience with the process. A regular voter might vote at most one to three times a year.
  A non–regular voter might vote once every four to eight years.

Typical polling place problems are these: 

• First, the voter has to wait too long to vote.
• Second, the voter goes to the wrong polling place or ends up in the wrong precinct within a multi-precinct polling site.
• Third, the voter is not on the registration list.
• Fourth, the election authority has difficulty in recruiting poll workers.

We address some tactics that should be explored by local election officials, to improve polling place practices respecting (1) the arrival process, (2) authorization to vote, (3) voter education, (4) staffing practices, and (5) continuous improvement.

A Provocative Scenario: It is 2002, because of long lines at the polling places in a major city, the city election office decides to leave the polling places open one extra hour. You, the reader, are standing in that line. The state courts close the polling places. You are denied access to vote.

Arrival Process

• First, make sure the voter knows where his or her polling place is located. Send cards to voters that indicate their registration status and where they go to vote. There could also be a public information campaign to let people know when they should expect a registration card and what to do if they do not receive one. Furthermore, an incentive can be given for using registration cards—you go to a shorter queue if you bring your registration card. The county could post on the Web information on where one votes.
• Second, at the polling place, make sure the voter knows where to go. Polling places with multiple precincts need to have clear signs indicating where the voter should go; for instance, at each entry there should be a map indicating the precincts and directions for which line to join.
• Third, encourage voters to vote during off-peak hours. Voters should be informed prior to the election about anticipated congestion during peak hours. On Election Day, real-time information on waiting times at the polls could be reported throughout the day, on radio, TV and the Web.
• Fourth, encourage early voting, as available. This also will help alleviate congestion during the traditional Election Day.

Authorization to Vote

• First, make sure the registration lists are as useful as possible. The various automated practices that update the registration lists should be examined, as we discussed previously in “Registration.”
• Second, try to deal with registration problems locally. Provide as much information as possible to each polling place. For instance, suppose that each precinct had a lap top computer with the entire  city’s registration list on it, and/or computer network access to the central files. The local precinct should be able to quickly tell an individual whether or not they are on the city’s list and where they may vote.
• Third, use provisional ballots. Determining the registration status or legitimacy of some voters may require an investigation. Such an investigation is potentially time consuming and very disruptive to do in real time, perhaps slowing the whole system down and impacting many other voters. The official in charge of a polling place should have the option to let the voter in question vote with a secure provisional ballot, but in such a way that the registration status can be investigated and resolved after the polls close.

Voter Education

• First, make sure the voter is prepared before entering the booth. Some fraction of voters will arrive at the polls not knowing for whom or what they wish to vote nor how to vote. A sample ballot should be printed in newspapers and made available on the Web prior to Election Day. In cities with many different ballots this may be difficult, but publishing a full-page sample ballot would probably help many voters. Some jurisdictions now maintain Web pages that allow voters to type in their addresses and see where their precinct is located. Including a sample ballot tailored to each precinct could enhance such database applications. To address voters who come unprepared for deciding for whom to vote, efforts should be taken to make voters familiar with the ballot and ballot format. Mock ballots should be available at the polling place for voters to review prior to voting. The voters should be encouraged to review ballot questions before entering the voting booth. Possibly they could be given an incentive to prepare a mock ballot before going to the booth. Laws that prohibit voters from taking mock ballots and similar aids into polling booths should be reexamined.
  To address voters who are unprepared for how to vote, the polling place must provide just-in-time training and education on an as needed basis. The polling place should have provisions in hand to identify such voters and to provide the requisite training quickly and without disruption to the rest of the flow. For instance, any voter who asks for help could be brought to an instructional area, where the voter can learn the steps required to execute a vote. This may require hiring additional poll workers at peak times.
• Second, keep everything as simple, visual, and selfexplanatory as possible. Given the limited experience most people have with voting, there is not much opportunity for familiarization or learning, so all operations and processes should be as simple as possible, requiring minimal explanation.
• Third, maintain continuity. For the same reason as above, changes in process should only be made when they will result in a long-lasting operational improvement.
• Fourth, train poll workers to be service agents for the voters. The poll workers must view their job as being there to assist voters to cast their votes accurately, with privacy and dignity. They need to be able to identify voters who might need help, and to deliver this help with respect and efficiency. 

Staffing Practices

• First, consider multiple shifts for precinct workers. The current practice is to have one shift, often running from 6:00 a.m. to 9:00 p.m. or 10:00 p.m. The average age of poll workers is beyond 65. Not surprisingly, it is hard to recruit workers.
  One might split the day into two eighthour shifts. This requires finding and training twice the number of poll workers.
  But it should expand the pool from which to draw and should result in a more alert work force.
  One might also consider shorter shifts to handle the peak hours, as is done with many service operations. For instance, there could be a three or four hour shift for the early evening. This would provide flexibility for varying the staff level over the day, corresponding to how arrivals vary over the day.
• Second, expand the pool of potential precinct workers.
  State laws often require precinct workers to be registered voters within the precinct. Such requirements should be reexamined. One promising source of potential precinct workers may not be eligible to vote at all—high school students. Some states already allow high school students to staff polling sites, through their community service programs. Such programs may not only help address short term staffing problems at polls, but may engage young people in the electoral process at an earlier age. 

Continuous Improvement

Every polling place should collect data on its operations so as to assess its performance and identify opportunities for improvement. For instance, a polling place might collect data on arrival of voters over the course of the day, waiting times, time to cast a vote, complaints, number of voters requesting help or education, registration problems and how they were resolved, and spoiled ballots.

Absentee and Early Voting

RECOMMENDATION

• Restrict or abolish on-demand absentee voting in favor of in-person early voting.

• Second, establish uniform reporting of absentee and precinct voting results.

Nationwide, fourteen percent of ballots in 2000 were cast outside of traditional polling places, either through absentee ballots or early voting. This contrasts with 1972, in which no state allowed early voting and only four percent of voters cast an absentee ballot. The 2000 election witnessed the first instance of a state (Oregon) conducting its presidential election solely by mail. In six other states (Arizona, Colorado, Nevada, Tennessee, Texas, and Washington), the fraction of ballots cast before Election Day (by absentee or early voting) exceeded twenty-five percent.

The most important formal features of absentee ballots is that they are generally cast before Election Day and delivered to the local election authorities by mail. Originally, absentee ballots could be requested only for cause. This is still true in most states. Justifiable causes typically include travel outside the voting jurisdiction on Election Day, service in the armed forces, illness or disability, and religious restrictions.

Over the past quarter century, many states have relaxed access to absentee ballots, allowing absentee ballots to be issued on demand. One example is California. Since 1978, any registered voter may apply for an absentee ballot between seven and twenty-nine days before an election, for any reason, including simple convenience. In 2000, nearly a quarter of California’s general election ballots were cast absentee.

In the 1970s and 1980s, states began experimenting with new types of voting away from neighborhood precincts. These modes formally share many characteristics with absentee balloting, but have been implemented for new reasons: namely, for the convenience of local residents who are not out of town on Election Day. These techniques are mail voting and early voting.

Voting by mail can be thought of as making mail-in absentee voting mandatory.
It operates at the initiative of election officials, who mail ballots to all registered voters, who then return the ballots to the court house, most often by mail. Nevada is credited with conducting the first election solely by mail in 1960 local elections. California followed suit; San Diego held the first large-scale election, a local referendum, entirely by mail in 1981.

Soon thereafter Oregon adopted a law allowing voteby-mail, first covering only local elections. In 2000, Oregon became the first state to conduct its general election entirely by mail. Currently at least sixteen states (Alaska, California, Colorado, Florida, Kansas, Minnesota, Missouri, Montana, Nebraska, Nevada, New Mexico, New York, North Dakota, Oregon, Utah, and Washington) allow vote-by-mail in at least some elections, although no other state has moved
nearly as far as Oregon.

Early voting can be thought of as stretching Election Day into an Election Period. States that have adopted early voting provisions generally make their election ballots available to all registered voters a couple of weeks before Election Day. How and where votes are cast varies. Most states allow voters to travel to the county courthouse to vote in person, regardless of where their neighborhood precinct is located. A few states, notably Texas, allow the establishment of satellite voting sites in government buildings and public places like shopping malls. States with early voting provisions in 2000 included Alaska, Arkansas, Colorado, Kansas, Nevada, New Mexico, North Carolina, Oklahoma, Oregon, Tennessee, and Texas.

Voting absentee, early, and by mail have grown steadily since the early 1970s, accelerating their growth since the mid-1990s. Voting away from neighborhood precincts has also tended to be more of a Western phenomenon. (See Map 2.) However, legislative changes and the strategic activities of political parties have also led to an eastward spread in non-precinct voting.

MAP 2

What Could Be Gained by These Techniques?

Arguments in favor of these three forms of voting all share a set of common aims.

• First, they are all intended to improve convenience for established voters. Established voters, who are the constituents of  most election officials, live busy lives and have experienced service improvements in the private sector affecting operating hours and procedures to accommodate complications of modern life.
• Second, because these techniques make voting more convenient, supporters contend that they should also decrease barriers to participation that confront current non-voters. With the barriers to participation lowered, voting turnout should increase.
• Third, many (though not all) of these techniques hold out the promise of reduced costs, at least in the long run. All-mail elections particularly eliminate the need to staff thousands of neighborhood polling stations, with their expensive equipment and staffing headaches.

Finally, all of these techniques promise greater administrative control over elections—not because they simplify elections per se, but because they provide more time for election administrators to handle the increasingly complex problems that arise in running elections.

What are the Dangers of These Techniques?

Five dangers are usually cited in opposition to early, absentee, and mail voting. 

• The first is coercion. The two primarily mail-in techniques (absentee voting and all-mail voting) are fundamentally not secret ballots. Although laws prohibit coercing absentee voters, the physical protections against coercion that exist in a neighborhood precinct—such as secrecy booths and buffer zones around polling places—are lacking.
  Concerns over coercion are especially acute in settings where voters may be reliant on care givers, as in nursing homes.
• A second concern is fraud and security. Mailing techniques rely on the delivery of ballots in unsecured modes. Mail channels to and from the court house are generally unsecured. The primary assurance that the intended voter returned a legal ballot is a signature on an affidavit that accompanies the returned ballot. Therefore, the integrity of the voting rolls depends on the signature verification skills of local election officials.
• The third concern is accuracy. Problems that voters might have in using voting technologies in precincts may be exacerbated in the mail-in settings. For instance, when punch cards are used in the absentee setting, the punch card ballot is often attached to a Styrofoam backing. A paper booklet with candidates and issues accompanies the ballot. The voter notes the number next to the candidate or yes/no position, locates that number on the pre-scored ballot, and punches through the number. Going back-and-forth between the booklet and the ballot introduces even more opportunities for mis-marked ballots than when punch cards are used in the precinct. Punch cards are especially confusing when the number associated with a proposition clashes with the hole number on the ballot card.
• Speed is a fourth concern with non-precinct voting methods. As the fraction of votes cast via the absentee process grows, concerns over delays in counting absentee ballots have also grown. In Washington State in 2000, for instance, over half of all ballots were cast absentee. The slowness of the count in Washington meant that recounts in two very close races, U.S. Senate and Secretary of State, were not ordered until three weeks after Election Day. Recounts in less visible local races were similarly delayed. 
• The fifth and final concern is that these techniques all reduce or eliminate the ceremonial aspects of voting. 

A Provocative Scenario: It is 2002, and in a tight race for the U.S. House, a voter complains that she did not receive her absentee ballot. The town election official says that the citizen actually voted. An investigation reveals that an organization applied for and filled out hundreds of absentee ballots of people on the "inactive" registration list. The election ends up going to the courts.

How Have These Techniques Fared?

Available data and scholarly assessments support those who urge caution in expanding opportunities for voters to vote away from neighborhood precincts. This is particularly true of mail-in methods, both early and absentee voting.

There is no evidence that liberalizing absentee voting laws or enacting early or vote-by-mail schemes has increased voter turnout dramatically. Oregon is a case in point. Oregon’s turnout in 2000—the first year of vote-by-mail for the general election—measured as a percentage of the voting age population, was up 3.5 percent over 1996, compared to the nationwide increase in turnout, which was 2.1 percent. However, sixteen states and the District of Columbia had turnout increases in 2000 that exceeded Oregon’s.

The story is less favorable in Texas. In every presidential election year since Texas began early voting in 1988, the voting turnout increase in Texas has been less than turnout increases nationwide. Early voting in Texas has therefore been associated with a net decrease in voter turnout, compared to the nation.

Research at the University of Michigan has documented that the most important effect of the Oregon vote-by-mail system has been to increase the convenience to established voters, not to induce many non-voters to the polls. Similar results have followed from research on Texas early voting and absentee voting generally. The one exception may be turnout in local elections. 

Research on the turnout effects of absentee voting are especially troubling in light of controversies in three Florida counties in 2000 over partisan use of ambiguous absentee voting laws.

There, lawsuits were filed in Bay, Martin, and Seminole counties, alleging irregularities with absentee ballot applications that party activists had sent to masses of voters. Leaving legal issues aside, the Florida episode reminds us that research has identified one condition under which absentee ballot laws increase turnout: when they are sufficiently ambiguous or liberal to allow partisan forces to use them to boost the turnout of party loyalists.

Lacking widespread and consistent data about electoral administration, it is difficult to document whether other gains from out-of-precinct voting have in fact materialized. For instance, the claim that all of these techniques provide a more manageable environment for dealing with the complexities of election laws seems true on its face. Yet voting jurisdictions rarely report reliable cost data. They also rarely report data such as the percentage of absentee ballots rejected due to irregularities. Without data such as these, assessing the administrative effectiveness of these techniques in new settings is virtually impossible.

A lack of data also impedes understanding whether current voting technologies are more or less errorprone in these settings. The spotty evidence that exists is inconsistent. For instance, in Idaho in 2000, the residual vote rate for absentee ballots was substantially higher among counties with punch cards (3.0 percent in-precinct vs. 4.6 percent absentee) while being roughly the same among counties with optical scanners (3.9 percent vs. 4.0 percent) and paper ballots (2.8 percent vs. 3.0 percent). In Florida, counties that separately reported election returns for absentee ballots generally showed no difference in the residual vote rate between absentee and in-precinct ballots.

Likewise, in Washington State the residual vote rate is unrelated to the fraction of ballots cast in counties that are absentee, once the size of the county is controlled for. 

In New York, on the other hand, the residual vote rate in 2000 for absentee ballots among a sample of counties was 4.4 percent, compared to the residual vote rate in those counties of 0.9 percent on the in-precinct lever machines. When Oregon instituted statewide vote-bymail in 2000, the residual vote rate went up statewide slightly compared to 1996, but the increase was significantly higher in counties that relied on punch cards (2.0 percent in 1996 vs. 2.3 percent in 2000) compared to counties with optically scanned ballots (1.2 percent in both years).

We performed a simple analysis to see if there was any correlation between the rate of absentee voting in counties and the rate of uncounted, unmarked, and spoiled ballots in 2000. The correlation was slight, and negative.

Speed of reporting is another concern that has arisen as absentee laws have become liberalized. Mail delays have always been a problem with absentee voting procedures. 

When the fraction of ballots cast absentee is small, these delays rarely have significant consequences. However, as the fraction of votes that are absentee grows, concerns over delays in counting ballots also grows. Speed of the count is a dimension on which Oregon’s vote-by-mail system offers clear advantages. By mailing out ballots to all voters and requiring that they be returned (by mail or in person) by 8:00 p.m. on Election Day, the Oregon system eliminates the need to wait for absentee ballots to trickle in after Election Day.
In short, absentee voting systems do not seem to have made great improvements in turnout. But they have also not produced higher residual vote rates.

The most important concerns raised by these procedures focus on increased opportunities for corruption. Indeed, the most prominent recent election fraud court cases involved absentee ballots— Dodge County, Georgia in 1996 and Miami in 1997. Dodge County involved two competing candidates for the Democratic nomination for the county commission bidding against each other for absentee ballots inside the county courthouse. In Miami, fraud so pervaded the absentee ballots that an appellate court eventually threw out all absentee ballots and declared a winner based solely on the machine vote.

We have no systematic measures of fraud, but fraud appears to be especially difficult to regulate in absentee systems. In-precinct voting or “kiosk” voting is observable. Absentee voting is not. The prospect for coercion is increased with absentee voting on demand.

Recommendations

• First, restrict or abolish on-demand absentee voting in favor of in-person early voting. The convenience that on-demand absentees produces is bought at a significant cost to the real and perceived integrity of the voting process. On the face of it, early voting can provide nearly equal convenience with significantly greater controls against fraud and coercion. Traditional absentee procedures for cause are still valuable for the limited situations they were originally intended for. States should return to those practices.
• Second, establish uniform reporting of absentee and precinct voting results. States should require that election jurisdictions report, in a uniform manner, data necessary to diagnose the accuracy and efficient administration of non-precinct ballots, as well as data necessary to ensure citizens that such procedures are no less accurate, error-prone, or fraud-prone than inprecinct methods.

These data include (1) separate election returns by method of casting a ballot (e.g., inprecinct, absentee, early), (2) cost accounts associated with administering different modes of balloting, and (3) statistics concerning the number of challenges to ballots and the reasons for excluding ballots from counting. Clear reporting will allow states to assess the effectiveness of absentee and early voting and to identify potential problems and irregularities.

Ballot Security

RECOMMENDATION

• Move away from complex, monolithic machines.
• Make source code for all vote recording and vote counting processes open source and source code for the user interface proprietary.
• Make recording software openly auditable in the same mode that is used to conduct the counts.
• Adapt equipment so that voters can create a record of the vote that they can examine directly, and that can be used to audit equipment and elections.
• Conduct audits of votes and equipment, even without a recount.
• Design equipment that logs all events (votes, maintenance, etc.) that occur on the machine.
• Train election officials in the interior workings of their voting equipment.
• Delay Internet voting until suitable criteria for security are put in place.

Security is as important as reliability in guaranteeing the integrity of the voting process and public confidence in the system. People do not use things in which they have no confidence.
Losing confidence in elections means losing confidence in our system of government. Security systems maintain our confidence that elections work.
Voters should not have to worry about rigged machines or illegal voters.

Stolen ballots, illegal voters, and stuffed ballot boxes have long been concerns in the U.S. They were the basic tools of machine politics in the nineteenth century.
Cases of vote fraud persist to this day. Some of the more ingenious methods for defrauding elections and some of the more entertaining stories involve stolen ballots.

In Maine in 1998, two legislative aides pleaded guilty to breaking into a ballot storage area in the Maine State House and tampering with the ballots being stored pending a recount of two close elections for the state legislature. When ballot tampering of this sort is discovered, sometimes the only remedy if the tampering affected the outcome of the election is a “revote.” Revotes are a bad way of settling contested elections because the election is no longer the same.
For example, if a single seat determined control of the legislature, then the revote would be not just about this seat, but about which party would govern the legislature. Worse still are the cases of fraud that we never discover; then the election does not reflect the public will.

The Maine case had an ironic twist. The defendants did not get the count right.
Their efforts to alter the number of ballots in favor of Democratic candidates failed because the absentee ballots from these races had not yet been tallied.
Although this case of tampering had no effect on the outcome of the race, it did negatively affect the citizens of Maine’s confidence in their electoral process.

Large-scale fraud—involving many voters or significant changes in the final tally—is more important than small-scale fraud involving a handful of ballots or voters. Small-scale fraud amounts to grains of sand added to one side of a scale. The U.S. historically has had problems of largescale fraud. Machine politics in the nineteenth century involved coordinated, large-scale activities to alter vote tallies, to cast illegal votes, and to destroy ballots. Such coordinated activities could alter thousands of ballots.
Small-scale fraud is also a concern in close elections, as the example from Maine demonstrates.
We distinguish two broad types of security problems: manipulation of voters and tampering with the recording of votes and counting mechanisms.

Manipulation of voters encompasses a range of activities. Your vote should be your vote: it should not be coerced or corrupted by someone else. And, every person should count the same: people are not allowed more than one vote in our society. Someone, such as an employer or union official, might coerce the voter to vote a certain way. Someone might try to purchase a voter’s ballot. Someone might try to vote more than one time. Someone from outside the community might come to vote in the community. Such manipulations involve the individual who casts the ballot.

The U.S. has developed a set of procedures to prevent manipulation of voters. The most important are the secret ballot and voter registration. Also, coercion, vote buying, and other forms of manipulation of voters are felony crimes.

Secrecy and registration are not themselves technological solutions, but they do place important constraints on the development of voting equipment. The secret ballot is a particularly troubling constraint because it means that, at least with current equipment, voting is receipt-free.

Tampering with the mechanisms for recording and counting votes represents a second type of security problem. Votes might be stolen or destroyed. For example, it is easy to jam a voting machine so that the counter in the back of the machine does not register the votes cast for a particular candidate. Votes might also be added to the count—stuffing the ballot box.
Before the election, someone might “warm up the machine” by pulling the lever a few times. These problems do not involve the individual who casts the ballot, but someone else, such as a poll worker, an election officer, or a manufacturer.
Many solutions for tampering with the mechanisms for recording and counting votes involve technology. Technology typically raises barriers to a security breach. Increasingly secure ballot boxes and machines have been devised. A common way of increasing security involves multiple locks, with keys controlled by the election administrators. Many systems have redundant recordings of the vote. Some jurisdictions require that the system have an “audit trail,” a separate recording of each vote that can be used to audit the performance of the machine. In addition, there is now a protocol for testing the integrity of tabulation software, though this procedure is voluntary and requires only limited testing.

Beyond technological protections, election administration provides a variety of protections against security problems. Poll watchers, who represent parties or candidates, can observe the goings on at each polling place and report any problems. In many states, counting is conducted publicly, to guard against altered or irregular tallies. Canvassing boards in some states check counts. In many states, police are assigned to each polling place; state and local police often oversee the transportation of ballots.

A Provocative Scenario: A programmer at SlickVotingMachines Corp. adds malicious code to a DRE (Direct Recording Electronic device) machine for the California 2004 Presidential election, so that every fiftieth vote for a Republican candidate is changed to a vote for the corresponding Democratic candidate. This only happens when the machine is in “real” mode as opposed to “test”mode, so the election officials never discover the fraud during their testing. The electronic audit trail made by the DRE machine is also affected, so “recounts”never discover anything amiss.

The security system for voting that has evolved in the United States has several important strengths that must be preserved as new technology is developed and deployed.

• First, we have an open process. Anyone can observe the activities inside the polling place, so long as they are not disruptive and do not try to persuade people to vote for a particular candidate, party, or position while in the polling place. Poll watchers frequently catch problems. In the 2000 general election in Boston, poll watchers noticed that the precinct wardens incorrectly recorded the counts for ballot propositions. The error retrieved 30,000 votes for at least one ballot question. Outside of polling place operations, it is important to have other parts of the process as open as possible to catch problems ranging from the design of equipment to the purging of registration lists, to the certification of the vote.
  Openness needs to be preserved, and this principle should be embraced throughout the voting system, including in the development of equipment.
• Second, the process involves many people and many different interests and separation of privilege, or separation of duty. The local election office manages equipment and counting of votes. However, within each locale there are many people watching the casting and counting of votes. The state governments certify the votes, and often oversee recounts. The vendors make the equipment, and have a stake in making sure that no one tampers with their equipment. No one person or level of government controls all elections.
  In fact, this is one important advantage to decentralization. Having many people observing each other provides an excellent set of checks.
  Decentralization guarantees many eyes on the process. We need to keep as many people and organizations as possible involved in the administration of elections and in the management and development of equipment.
  We should not give an individual or computer system more control over the process than he, she, or it needs to do his, her, or its specific function.
• Third, most equipment does, and all equipment should, provide redundant trusted recordings. Having several recordings of voters’ intentions allows a full audit of any election. Typically we only audit close elections, when a recount has been requested. Some jurisdictions and some vendors will audit equipment at random to learn more about what works and what does not, and to try to identify problems that might exist throughout the system. For reasons of security, we should require redundant recordings of all new equipment, and we should move away from equipment that does not allow for redundancy.
• Fourth, election administration is a public process. U.S. elections are administered by public officials rather than by private agencies. Because elected officials are ultimately responsible to the public, we the voters have a higher degree of control over their performance.
  Because voting is a public good, public control is essential.

We strongly believe that election officials should have full control over all equipment used in elections. They may contract out for service and storage and even lease equipment, but election officials must be able to inspect all aspects of equipment at any time.

Electronic Voting and Security

We are concerned that we are moving away from these general principles that help guarantee the security and integrity of voting.
We are in an era of electronic voting. Almost two thirds of all votes in 2000 were counted using electronic tabulation, including computers, punch cards, scanners, and DREs (Direct Recording Electronic devices). Hand-counted paper, despite its advantages and wide use in Europe, is infrequently used in the U.S.
Electronics are increasingly used to record votes. DRE machines require that voters generate votes and record votes electronically. Scanners and DREs are where the growth in the industry is occurring.
The computerization of election systems introduces significant security risks but also significant opportunities for fraud prevention and detection. For example, electronic transmission of vote tallies, so long as that transmission is secure, means that we do not have to have police ferrying ballots around on election night.
We see the following security risks associated with electronic voting.

• First, we are losing openness. Electronic voting machines are completely closed. We can no longer observe the count.
• Second, we are losing the ability for many people to be involved. Election equipment tries to do it all. A single
  computer system generates votes, records votes, counts votes, and produces the final tallies. Without openness, we lose the advantage of having many eyes on the count.
• Third, separation of privilege is lost. We are headed toward monolithic systems—one machine that does it all. This risks vesting too much control over the system in the vendor’s hands or in the hands of any hacker who can get inside of that monolithic system.
• Fourth, many electronic devices lack redundancy and true auditability. To audit a voting machine, one needs a redundant recording of what the voter intended.
  There is the initial recording that the electronic machine made, but there must also be a separate recording against which the machine recording is tested —an audit trail. The problem for many electronic devices is that their audit trails are simply another recording of what the machine recorded. Roy Saltman, a leading expert on voting technology, has long advocated that the true standard of auditability is that the audit trail is produced by the voter and not by some intermediary machine. This is an important insight. It is the only way to guard against a fraud scheme in which the code occasionally drops votes; it also protects against machines that accidentally lose votes, say because of a power surge.
• Fifth, we are losing public control over voting equipment. One worry with electronics is that they are sufficiently
  complex machines that administrators cannot inspect the inside of the devices. Even the independent testing authorities have difficulty completing speedy certification reviews of the hardware and software on new electronic devices owing to the increased complexity of the hardware and software.
  Administrators must trust manufacturers, as must the voters. We prefer transparent voting systems where the operations are observable and verifiable by anyone.

All of these problems are solvable. We strongly believe that the principles of openness, many eyes, separation of privilege, redundancy, and public control must guide the design of electronic equipment.

• First, we should move away from complex, monolithic machines. It is very difficult to design secure systems that must meet a complex set of requirements. Extreme simplicity is strongly recommended. We think that a better approach is to have a very simple electronic vote-recording device that is separate from other parts of the system. A machine used to prepare a ballot can be as complicated as one likes, and could even be used for other things when elections are not happening, such as classroom instruction.
  The vote-recoding device is the critical device in securing the vote. When the vote is recorded is the moment that the voter loses control over the vote. All of the problems of tampering emerge at this moment.
  If the vote recording is secure, then we can truly heighten the security of the entire system.
  What must be secure are the devices that record and count, not the user interface that generates ballots.
  The device that records votes must be very secure.
  And it should not be expected to do anything other than record votes. It should be a very simple machine, nothing as complicated as a personal computer. This suggests that the industry and administrators use separate devices for recording and for generating votes. That will be explored in the final section of this report.
• Second, the source code for all vote recording and vote counting processes must be open source. The source code for the user interface can and should be proprietary, so that vendors can develop their products.
  There are many protocols for open source. We think that a national commission consisting of experts on security from outside the voting industry, including other industries such as banking and Internet security, should determine the appropriate protocol for open source in the voting equipment industry.
• Third, all recording software should be openly audited in the same mode that is used to conduct the counts.
  “Test” modes should be eliminated. Counting and recording devices should be “modeless.” The test mode feature is a security vulnerability because it creates a way to cover a hack. To truly reclaim the openness of the count, interested parties (candidates, party organization, groups, etc.) should be allowed to inspect the software as it is formatted for Election Day. All interested parties should be satisfied that votes will be counted appropriately.
• Fourth, equipment should be adapted so that voters can create a record of the vote that they can examine directly, for the sake of auditing equipment and elections. This might require some sort of simple paper recording that the voter can check and submit separately.
• Fifth, we recommend audits of votes and equipment, even without a recount. Total votes and votes for each office and proposition should be logged on all equipment and recorded electronically. Election officials should inspect these recordings to detect irregularities on particular machines or at particular precincts. In addition, election officers, especially in larger jurisdictions, should randomly choose a small percent of the machines (say one percent) each year for thorough inspection.
• Sixth, all equipment should log all events (votes, maintenance, etc.) that occur on the machine.
  The information on the log should include what was done, when it was done, and who authorized the activity. The election office should keep those logs.
• Seventh, all election officials should be trained in the interior workings of their voting equipment. They should only use what they can understand and check.
  This training is perhaps best provided by vendors, and may be a requirement of purchase.

Finally, we are concerned with where computerized voting is heading. Voting on a personal computer is a step away from voting on the Internet. Remote Internet voting poses serious security risks. It is much too easy for one individual to disrupt an entire election and commit large-scale fraud.

Cost and Public Finance of Elections

RECOMMENDATION

• The federal government, working with state and local governments, needs to develop standard methods of accounting for election expenses and standard reports that are made publicly available.

• The federal and state governments should offer significant matching funds for upgrades to replace voting technologies—such as punch cards, lever machines, centrally counted optical scanning, paper, and some
  under-performing DREs—that are clearly dominated by existing equipment.

• The federal and state governments should pay for the maintenance of voter registration databases maintained at the state and local levels.

• The federal government needs to maintain a publicly available database of election expenditures.

Elections represent an organizational challenge in a country like the United States, with nearly one quarter of a billion eligible voters scattered across over 3,100 counties in fifty states. To meet this challenge we have developed a large bureaucracy—or rather, thousands of small bureaucracies. A small but vital group of private vendors and service providers produce the equipment, software, and
peripherals to collect, process and count millions of individually marked ballots.
These firms also develop software for voter registration and other aspects of election administration.
We refer to this public/private partnership of election producers collectively as the election industry.

This industry produces a service. Do we spend an appropriate amount for this service? Or, as is common with public goods, do we purchase too little service in support of elections?

One answer is that we should spend more. Elections are fundamental to our society, and the U.S. promotes democracy around the world. The operation of elections in the U.S. should be given a very high priority. Today, elections receive about as low a priority as any government service.

Perhaps this is answer enough—the U.S. should change its priorities.

We do not give elections a high priority, and we must consider how this industry manages to provide the services that it does under existing financial constraints. Within these constraints, are improvements possible?

Here we take a harder and more analytical look at how we provide for elections in America.

Even the most basic facts about the cost and finance of elections in the United States are unavailable, and the most basic questions remain unexamined. It is not known how much we spend on election administration overall in the U.S. each year. It is not known on what funds are spent. There has been little analysis of how and how well local governments provide election services. Each of us has some sense of what we get—a stable and successful democracy.
But there are clearly problems that can be remedied. How much will improvements in this system cost?

In preparing this report, we have collected data to try to assess some of the fundamental questions. How much do we spend on elections, and on what? To make decisions about the value of additional expenditures, more thorough study will be needed.

Inputs and Outputs of Elections

What is output? As is often the case in service industries, including the delivery of election services, this is a bit slippery. Output can be defined in several ways: customers (voters) served, the functioning of the system supported (democracy), and many measures in between.
Even among services, the output of the election industry is unusual for a number of reasons, three of which are particularly noteworthy.

First, it is one of a handful of industries that is financed entirely with public funds. Remarkably, the public finance issues have never been investigated systematically.

Second, the output, valid ballots, is unpriced and untraded in the marketplace (or at least that is supposed to be the case). Moreover, it is difficult to place an economic value on valid ballots, much like valuing the protection of an endangered species or a national treasure. It is a classic public good problem, and different individuals will place different values on it. The total value to society would require us somehow to aggregate these individual valuations.

Third, the output is also an intermediate product, with the “real” output being the electoral outcome (winning and losing candidates and propositions) and, ultimately, public policy.

By some measures we are doing impressively well. The electoral process has survived civil and international wars. It has expanded dramatically to include many new categories of people over the past two hundred years without disrupting our form of government. In 2000 alone, over 100 million people voted.
Our report emphasizes a different output: the quality of service. And we have one such concrete measure: lost votes. We must lower the rate of lost votes to an acceptable level.
How much is it worth paying to have a marginally more honest and accurate election? We do not yet know the answer to this question. Instead, we can assess how much the existing level of quality (vote loss) costs.

To compute costs, we consider “What are inputs?” This is easy to answer in principle, but hard in practice, because of poor data. Basically, the inputs are labor, maintenance, storage, and acquisition of equipment, supplies (such as printing), information systems, and rental of space (often free). Cost figures are available usually at the county level in one of two forms: annual election budgets and general election operating costs.

A common measure of election cost is given by costper-vote, or alternatively, cost-per-valid-ballot. This measure does not seem appropriate in many circumstances because of the varying number of ballot items, differential turnout rates (“potential votes”), and other quality issues. It is, however, the one we have the most information on, and it is useful so long as quality does not vary too much with the key independent variables (like machine type or number of ballot items).

Production costs for vendors are not available, and there are significant development and marketing costs that are bundled into the equipment contracts.

A Provocative Scenario: It is 2004, and little has been done to improve the voting system. A dip in the economy led to belt-tightening in state and local budgets. Education and roads were spared; new voting equipment and registration systems were put on hold. In close elections around the country, media scrutiny once again reveals that the problems of 2000 remain unfixed.

How Much Do Elections Cost in America?

Surprisingly, there is no ready answer to this question.

The reason? Election expenditures are sufficiently small that they do not make the list of important activities reported in the Census of Governments, which is the annual report by the U.S. Census Bureau about what states and local governments spend on their functions.
That, in itself, is some indication of the low level of election expenditures in the United States. The smallest general expenditure category presented by the Census of Government for the Statistical Abstract of the United States in 2000 is $14 billion on solid waste management.

Accounting practices also contribute to the difficulty of measuring election administration expenditures.
For example, some counties have very detailed budget reports, including space rental, printing costs, telephone and postage, pollworkers, etc., while other counties offer no budget breakdown at all.

Overall Spending

We canvassed county and state governments from around the country to find out how much they spend annually on elections.
Based on annual budgets from various states, we estimate that the counties spent approximately $1 billion on election administration (excluding some large procurements of new equipment) in 2000. That was for a presidential election year,
so actual operating costs are somewhat lower in other years. With slightly over one hundred million voters, that works out to about $10/voter. States and cities also contribute to the financing of elections, but at a lower level than the counties.
We sought other estimates to corroborate ours and found them to be in the same neighborhood—approximately $1 billion, perhaps slightly more. For instance, an analysis of election administration in California (commissioned by Ernest Hawkins, Registrar of Voters of Sacramento County, California) arrived at a similar projection for the U.S.

A much more in-depth census of election administration is required to give a complete picture of what is spent on election administration. We take the $1 billion figure to be a ballpark estimate. To put it in some context, counties and cities spend over ten times that amount on solid waste management and on parks and recreation.

How much is spent also depends on the size of the county. There are some economies to scale in election administration, as indicated in Figure 1, which is based on data from a survey of annual election budgets from counties in nine states in 2000.

Very small counties (less than 25,000 voters) spend disproportionately more to run their elections. This is probably not because they have more resources to spend: rural counties tend to be poorer. Above 25,000 voters, there is little evidence of an economy to scale.

Itemized Expenditures

Far more difficult is figuring out how much we spend on specific aspects of election administration—equipment, voter registration, polling place operations, and other factors. By looking at detailed breakdowns of the budgets of several cities and counties, we have been able to approximate the division of these costs into equipment purchases and maintenance, Election Day operations, voter registration, and general administration.

Voter registration and general administration account for the lion’s share of election expenditures—roughly one-third each. That is, counties and local governments spend between $300 million and $400 million each year on their registration systems. We have learned from the voting equipment industry and from local budgets that equipment purchases and maintenance amount to approximately $150 million to $200 million annually, or roughly fifteen to twenty percent of total election administration expenditures. Election Day operations—polling place management, poll worker training and salaries, printing, and the like—are in a similar range, fifteen to twenty percent of total budgets.

On a per voter basis, these figures imply that locales spend approximately $3.50 per voter on voter registration.
Local governments spend approximately $1.50 per voter to acquire and maintain voting equipment. And local governments spend about $1.50 per voter actually to run the election on Election Day. Another $3.50 per voter is spent on administrative overhead.

Economies to scale are much more evident in the operation of elections. These scale economies arise not just because of a large fixed cost of equipment or administration.
Larger counties also spend less on a per voter basis for Election Day operations. Consider the operating expenses for the 2000 general election of the counties in North Dakota. Figure 2 below graphs county size on the horizontal axis and per-voter general election costs on the vertical axis. On a per voter basis, smaller counties pay much more for Election Day operations than large counties.

Equipment Expenditures

Equipment costs are of particular concern. Today, many states and counties wish to upgrade their equipment.
Some states are forcing counties to upgrade through legislation banning or decertifying punch cards. What is the fiscal impact of such purchases?
We have obtained information from some counties and vendors on the costs of acquiring equipment. There are two competing technologies—direct recording electronic machines (DREs) and optical scanners.

Acquisition costs for purchasing new voting equipment are $18–$25/voter for touchscreen systems and $8–10 for in-precinct optical scanning equipment.
The lifespan of election equipment ranges widely, but averages in the 15–20 year range, so that acquisition costs of current equipment would be on the order of $1–$2 per voter per year in current dollars.

A nation-wide upgrade to touchscreen DREs would cost up to $2.6 billion; a complete upgrade to scanners would cost up to $1 billion. Both figures are well in excess of what all counties currently budget for equipment (approximately $150 million).

These figures, however, seem less exceptional when we consider the lifetime of the equipment. Assuming these machines last approximately 15 years, the cost of an upgrade to DREs would run approximately $1.40 per voter per year and the cost of an upgrade to scanners would cost approximately $0.60 per voter per year. This is well within the annual revenues generated by equipment sales. Industry sources also report that annual industry revenues are on the order of $150 million in a good year. This corresponds to $1.40 per voter per year.

This figure is in line with what counties currently spend on equipment, approximately $1.50 per voter per year. However, most counties bear the costs of purchases at one time. A purchase of DREs of $20 per voter is double the typical county’s entire election administration budget.

The fiscal problem is figuring out how to finance equipment purchases over the long run.
One solution is that the states or federal government share the cost of any changes in equipment, especially if the states or federal government mandate changes. Counties cannot deal with unfunded mandates of this magnitude.Another part of the solution is leasing. The state of Rhode Island, many counties in Maryland, and a scattering of counties elsewhere lease equipment. Leasing avoids the huge upfront expenditure for purchasing equipment, and leads to greater flexibility for upgrades.

Leasing contracts—including maintenance, service, and consulting—are on the order of $1.50 per voter per year (over fifteen years), based on Rhode Island’s lease-to-own agreement with Election Systems and Software (ES&S). The state has approximately 400,000 registered voters and the annual cost of the contract is approximately $0.6 million.

The Maryland Secretary of State’s office recently published a report that showed leasing costs ranging from $1 to $3 per voter per year, with much of the variation attributable to differences in population density. These figures are only slightly above what other counties budget for equipment maintenance and purchases annually.

There may be some premium for leasing—that is, leasing election equipment may cost more than buying it, in the long run. However, states and counties can strike lease agreements that actually lower costs.
Rhode Island’s state legislature stipulated that the lease could only happen if the total cost (including service and equipment) cost less each year than maintaining the state’s lever machines.

How Much Quality For Each Dollar?

Election officials like to point out a “tradeoff triangle” that reflects what many administrators and the news media view as the primary three objectives of a “good” election system. These are: (1) speed of the count, (2) accuracy, and (3) cost. Note that this ignores other criteria, such as security, ease-of-use, and accessibility.
In our view, one of these—counting speed—is misplaced.

Surely this must be a secondary consideration given that the outcomes of most elections are not implemented for months. Moreover, virtually all systems have counting speeds that enable the election to be called within twenty-four hours, barring recounts or unusually close elections. In the rare cases where twentyfour hours is not sufficient, most experts would agree that it could not be done in fewer than twenty-four hours using any of the existing technologies, except
perhaps those that effectively preclude any hand recount—systems without a paper trail, such as lever machines and most electronic devices. Counting speed, therefore, seems nearly irrelevant.

Perhaps more critical is recounting speed. Long recounts damage public confidence in the election system and open up greater opportunity for fraud.
Thus, the main tradeoff is between cost and accuracy.

The two main contenders in terms of modern technology are precinct-level optical scanning and touchscreen electronic. Over a fifteen-year span, the combined operating and acquisition costs are not substantially different. Both are around $2 per voter per year.

The additional annualized cost (or savings) from choosing electronic instead of optical scanning would not be more than 10 percent of the total annual election administration costs. Thus, equipment costs are at a very reasonable level, with only marginal variations across the two prime technologies of today.

Given our most current data, there is a difference between electronic voting systems and precinct level optical scanning technology. Optical scanning has produced significantly fewer residual votes than electronics over the last decade. However, this is based primarily on data from full-face DRE equipment.
Touchscreen equipment is very new and has a limited track record. As the industry improves electronic technology, the gap in the residual vote rate between these two technologies can be expected to narrow.

The Future of Voting Equipment Manufacturing

By modern standards, the voting equipment industry is small. A small number of private firms invent, develop, manufacture, market, and maintain voting equipment and election supplies for the counties. Industry estimates of annual industry revenue fall in the $150–$200 million range, or about $1 per eligible voter. This figure covers sales of new equipment, maintenance, and service (including printing of ballots, in some cases). To put this in perspective, annual sales of residential lawnmowers run into the billions of dollars, making the residential lawnmower industry more than ten times the size of the entire election industry.

In the past decade, companies involved in elections have undergone a major consolidation, leading to a more concentrated industry. The four largest manufacturers are Danaher Controls (Guardian Voting Systems), Global Election Systems, Election Systems and Software (ES&S), and Sequoia-Pacific Voting Systems. Together, they make up nearly ninety percent of the market. By far the largest of these is ES&S, which contracts with approximately sixty percent of
the counties in the U.S. Very few counties contract with more than one vendor.

Because of the long shelf life of the product—twenty years or more—relationships between a county and its vendor are long-term. Contracts are negotiated each time a new equipment purchase is made, often between savvy veterans from the company sales force and county officials who rarely, if ever, negotiate any major contracts and are unlikely to have negotiated a previous contract for election equipment.

We do not expect much growth in this industry. Assuming that all counties upgrade their equipment over the next fifteen years and that one-half adopt DRE devices and one-half adopt scanners, we project that the industry will remain approximately the same size.

One perverse effect of the current push to purchase new equipment is that it may hasten the need to develop a new business model in order for firms to survive. Suppose that all counties with obsolete or inferior equipment upgrade within the coming year, so all counties have relatively new, relatively good equipment. This will kill demand over the succeeding years.

The next few years will likely be quite good for those selling machines, but the long-term prospects for this industry are not as rosy. Now is a critical juncture for firms to evaluate the service they provide and to make a serious effort to develop new ways of providing voting technology.

The voting equipment industry must adapt in order to thrive. A new business model might emphasize service over selling boxes. It might emphasize modular equipment with standard operating systems: one firm provides the machine, one firm provides the user interface, one firm provides the counting and vote transmission software.

A Federal Role in Financing Elections

The federal government and most state governments have stayed out of financing election administration.
We can identify several specific ways the federal government can contribute to the public finance of election administration.

The government should finance upgrades of equipment to phase out dominated technologies (punchcards, lever machines, centrally counted optical scanning, paper, and under-performing DREs). A preferred approach would involve a gradual and on-going process for administering grants to counties and localities to help them replace deficient technology in a methodical and carefully studied way that would create options for future system upgrades or conversions.

The federal government should establish and fund an independent agency for election administration.
Currently, there are significant financial constraints on the Office of Election Administration in the FEC.

The new agency would perform the sort of information clearinghouse function that we see as necessary in order to establish best practices and to improve the information that counties have when they purchase equipment. In addition, it would oversee federal grants to counties for voting equipment, grants to conduct research on voting equipment, and head up an office of standards and certification.

The agency should develop accounting standards for reporting election expenditures and equipment field performance. This needs to be done in order to assess the efficiency of different election systems, and to pinpoint the best places to invest resources for improved performance.

The federal government should provide research funding for the innovation and test-bedding of cutting-edge technologies. One possible way would be to establish a program to field test new technologies in a rigorous and carefully planned way. Without first conducting field pilot tests with real elections on a small scale, the implementation of these technologies is subject to risk.

The federal and state governments should finance and coordinate the upgrade and ongoing maintenance of voter registration databases for counties and states.
Voter registration is the largest component of election administration costs, accounting for expenditures of around $350 million a year. Funds should be available for counties on a per capita basis.

NEXT PAGE       CALTECH INDEX

Farming Politics Government Posters Humour Technology Religion Nature Me Links