Election technology: from lever machines to paper trails and voting system standards

Election technology is the equipment and software used to present ballots, record votes, and count them. In the United States it moved from hand-counted paper to lever machines, first used in 1892 in Lockport, New York, then to punch cards in the 1960s, optical scan ballots, and direct-recording electronic machines.⁠¹ The 2000 presidential recount exposed the limits of punch cards and led to the Help America Vote Act of 2002, which requires voting systems to produce “a permanent paper record with a manual audit capacity”.⁠^3,4 Security reviews and a 2018 National Academies report then pushed elections toward human-readable paper ballots.⁠^6,7

In plain words

For a long time Americans voted on paper and people counted the paper by hand. That was slow and open to cheating, so cities bought machines.

The first big machines used levers. You pulled a lever for each choice, and gears inside counted. They were hard to tamper with one ballot at a time, but they left no ballot to recount.

Next came punch cards. You poked holes in a card, and a computer read the holes. In 2000, Florida’s presidential recount hinged on bits of paper, called chads, that were only partly punched. Different counties judged them differently, and the Supreme Court ruled that the recount, as it was being run, treated ballots unequally.

In 2002 Congress passed a law setting standards for voting machines, including a paper record that can be audited. Some places used touchscreen machines. In 2007 a California security review found enough problems that the state pulled its approval of several systems and allowed them back only under stricter conditions.

So the field went back to paper, now read by scanners and checked by audits. The lesson: a voting system should leave proof that ordinary people can check, without having to trust a computer.

Defined terms

• Lever machine. A mechanical device in which voters pull levers and counters record totals. Douglas Jones’s history records the Myers Automatic Booth as first used in 1892.⁠¹
• Punch card ballot. A card on which a voter punches holes, read by a tabulator. The Votomatic was adapted for voting at the University of California, Berkeley, in the early 1960s.⁠¹
• Optical scan (mark-sense) ballot. A paper ballot marked by hand and read by a scanner. Jones dates the first use of mark-sense ballots to 1962 in Kern City, California.⁠¹
• DRE (direct-recording electronic). A machine, often a touchscreen, that records votes electronically. Jones identifies the Video Voter, first used in 1975 in Illinois, as probably the first DRE used in a real election.⁠¹
• VVPAT. A voter-verified paper audit trail printed by an electronic machine. See voter-verified paper record.
• Residual vote. A ballot with no countable vote for a given race, from undervote or overvote. It is used as a rough measure of lost votes.⁠^2,8
• VVSG. The Voluntary Voting System Guidelines adopted by the U.S. Election Assistance Commission.⁠⁵

Evidence

History: the mechanical era

Douglas W. Jones, a computer scientist at the University of Iowa, writes in A Brief Illustrated History of Voting that “the Myers Automatic Booth lever voting machines was first used in 1892 in Lockport, New York,” and that “by the 1930’s, essentially all of the nation’s larger urban centers had adopted lever voting machines”.⁠¹ Lever machines prevented some ballot box fraud because there was no loose paper to add or remove, but they also left no individual ballots for a recount.

History: punch cards and scanners

Jones records that “in the early 1960’s, two professors at the University of California at Berkeley adapted” IBM’s Port-A-Punch for voting; Joseph P. Harris and William Rouverol formed Harris Votomatic, Inc., which IBM bought in 1965 and left in 1969 after problems surfaced.⁠¹ Optical scanning was developed in the mid-1950s under E. F. Lindquist at the University of Iowa, and “the first use of mark-sense ballots was in 1962, in Kern City, California”.⁠¹

History: the 2000 recount

The 2000 presidential election in Florida brought punch card problems into national view. In Bush v. Gore, decided December 12, 2000, the Supreme Court described ballot cards where “a chad is hanging, say by two corners,” and noted that standards for reading ballots “might vary not only from county to county but indeed within a single county.” It held that “the want of those rules here has led to unequal evaluation of ballots,” a violation of equal protection. It also wrote that “our consideration is limited to the present circumstances”.⁠³ The opinion stated that “seven Justices of the Court agree that there are constitutional problems with the recount” and that “the only disagreement is as to the remedy”; Justices Stevens and Souter wrote dissents, joined in part by Justices Ginsburg and Breyer.⁠³ Ballot design problems in the same election, including Palm Beach County’s butterfly ballot, are covered in the Australian ballot in America.

Research results: how many votes were lost

The Caltech/MIT Voting Technology Project reported in July 2001 in Voting: What Is, What Could Be that “between 4 and 6 million votes were lost in the 2000 election.” It estimated that replacing punch cards and lever machines with optical scanners could recover about 1.5 million lost votes, and that better registration databases, polling place access to registration data, and provisional ballots could save another 3 million.⁠² Registration problems, in that estimate, cost more votes than machines did.

Residual votes are not purely a technology measure. Charles Stewart, R. Michael Alvarez, and colleagues, in Social Science Quarterly (published online 2019), found that the national residual vote rate for president rose from 0.99 percent in 2012 to 1.41 percent in 2016, mainly because of intentional abstention, and that the rise was not caused by changes in voting technology.⁠⁸

Law: HAVA’s equipment rules

The Help America Vote Act, signed October 29, 2002, sets minimum standards for voting systems in federal elections.⁠⁹ Its Section 301, codified at 52 U.S.C. 21081, requires that voters be able to verify and correct their selections privately before casting, that systems notify voters of overvotes, that systems produce “a permanent paper record with a manual audit capacity” that serves as the official record for recounts, that each polling place have at least one system accessible to voters with disabilities with equivalent privacy and independence, and that each state adopt uniform standards for what counts as a vote. States had to comply by January 1, 2006.⁠⁴ The uniform standard requirement answers the problem Bush v. Gore identified.

History: security reviews

In 2007 California Secretary of State Debra Bowen commissioned a “Top-to-Bottom Review” of certified systems, with University of California computer scientists and red team testers, to “restore the public’s confidence in the integrity of the electoral process.” On August 3, 2007, she withdrew approval of systems from Diebold (later Premier), Hart InterCivic, Sequoia, and Election Systems and Software and reapproved most under strengthened security conditions.⁠⁶ Specific vulnerabilities are documented in the linked technical reports.

Current standards: paper and national guidelines

In 2018 the National Academies of Sciences, Engineering, and Medicine, in Securing the Vote: Protecting American Democracy, recommended that elections “should be conducted using human-readable paper ballots by the 2020 presidential election,” and that marked ballots “should not be returned over the Internet or any network connected to it,” because no current technology can guarantee their secrecy, security, and verifiability.⁠⁷

The Election Assistance Commission adopted VVSG 1.0 in 2005, 1.1 in 2015, and 2.0 unanimously on February 10, 2021. Adherence is “voluntary except in select states” whose laws require it. Since November 16, 2023, new certifications must be to VVSG 2.0 or limited modifications of existing systems, though older systems are not decertified.⁠⁵ Paper ballots make risk-limiting audits possible; see the Colorado case and audits and verification.

Competing interpretations

Technology as accuracy. One reading holds that better machines reduce lost votes, as the Voting Technology Project’s estimates for replacing punch cards and levers suggest.⁠² Electronic machines also enable accessible voting, which HAVA requires.⁠⁴

Technology as risk. The security reading, reflected in the California review and the National Academies report, holds that software-dependent systems can fail or be attacked in ways that leave no trace, so the record of the vote should be paper a voter has seen.⁠^6,7 Supporters of electronic systems answer that paper has its own failure modes, from lost boxes to ambiguous marks, as 2000 showed.⁠³

Trust versus verification. Some argue that confidence comes from certification by experts and testing labs.⁠⁵ Others argue that confidence should come from evidence any observer can check, such as audits of paper. These are not exclusive, but they assign trust differently: one to institutions, the other to evidence.

Equal treatment versus local control. Bush v. Gore found unequal ballot standards unconstitutional in that case while limiting its holding.⁠³ HAVA required uniform standards within each state.⁠⁴ Election administration in the United States remains largely local, and debates continue over how much uniformity is needed.

Interpretation, this site’s reading. The history supports a principle more than a device: evidence first. The best equipment is whatever leaves a record the public can use to confirm the count without trusting the equipment.

Application to this framework

These are design implications, not claims about the framework’s performance.

1. The framework conducts no official votes. My Vote Forward is educational, its demo uses simulated data, and its software is not a voting system under HAVA or VVSG. EAC guidelines are not approvals of this project. See is this an official voting service.
2. Evidence over assurance. Any future private voting component would need independently checkable evidence, such as paper records or end-to-end verifiable cryptography, plus audits. See private voting and security.
3. Internet ballot return is not solved. The National Academies’ 2018 recommendation against returning marked ballots over the internet applies directly.⁠⁷ See the Washington, D.C. 2010 test.
4. Define what counts in advance. HAVA’s uniform-definition rule.⁠⁴ has a parallel: rules for what counts as a valid priority submission should be fixed before collection.
5. Accessibility is built in. HAVA requires accessible systems with equivalent privacy⁠⁴ See disability access and hardware.

Hypothetical example. A town tests an app that lets residents rank budget priorities, then proposes using it for an official advisory vote. Even if the app works well, it produces no record that voters have seen and that observers can recount. Before any official use, the town would need certified equipment, a voter-verifiable record, and an audit plan set by the responsible election authority.

Unresolved questions

• Can ballot-marking devices, which print a paper ballot from a touchscreen, provide both accessibility and voter verification, given open questions about how often voters actually review the printout?
• Will any form of remote electronic ballot return ever meet the National Academies’ conditions of secrecy, security, and verifiability,⁠⁷?
• How should states that do not require VVSG compliance assure the public about equipment that has not been tested to current guidelines⁠⁵?
• How much did lost votes in 2000 come from equipment versus registration⁠² and has that balance changed since?