Join top executives in San Francisco on July 11-12, to hear how leaders are integrating and optimizing AI investments for success. Learn More

According to a Pew Research Center analysis, nearly half of registered U.S. voters live in jurisdictions that use optical scan machines as their voting system as opposed to touchscreens or other direct-recording electronic (DRE) devices. Another 19% live in counties where both optical-scan and DRE systems are deployed.

Speed and cost savings are often cited as advantages by the counties that have adopted optical scanning. But optical scanners are limited by the robustness of the algorithms they use. Poorly designed and uncalibrated machines could inadvertently miscount or reject ballots, in some cases without voters noticing.

Optical scanners

Optical scanners aren’t new technology. Mark recognition dates back to the early 1950s, when it was first explored in the context of standardized tests such as entrance exams. The Norden Electronic Vote Tallying System was the first ballot-counting scanner of its kind to market, but it required the use of special ink. Machines like the Votronic dispensed with proprietary ink in favor of pencil graphite.

Scanners today use digital imaging technology. Each voter’s choices are marked on one or more pieces of paper that pass through the scanner, which creates an electronic image of each ballot, interprets it, and tabulates the votes. The images are usually stored in databases for later review.


Transform 2023

Join us in San Francisco on July 11-12, where top executives will share how they have integrated and optimized AI investments for success and avoided common pitfalls.


Register Now

Digitization doesn’t correspond with sophistication, however. For example, the Avante Vote-Trakker simply counts the number of dark and light pixels in marking areas (like ovals) to determine if the mark is a vote. Other systems use algorithms sensitive to the shape of the mark as well as the overall darkness. Still, officials have identified hundreds of errors in optical scan systems, frequently caused by feeding ballots upside down; pulling multiple ballots through at once; paper jams; broken, blocked, or overheated sensors; printing that doesn’t align with the programming; and programming errors.

During the 2000 presidential election, Orange County, California’s optical scan machines failed to count more than 400 votes “for no apparent reason.” And tens of thousands of ballots in Florida weren’t counted as the result of scanner failures. Roughly 31,775 voters ignored, didn’t read, or couldn’t understand instructions to use a No. 2 pencil or blue-ink pen to mark their choices. They instead drew stars, circles, and Xs; used pens with ink invisible to scanners’ infrared sensors; and attempted to correct errors with tape and staples. Others made marks by a candidate’s name but then filled in a second bubble and wrote the same name in the space for write-in candidates. And some erroneously filled in ovals excepting those next to the candidates they intended to choose.

Voters in Napa County, California, encountered similar issues during the 2004 election. Scanners manufactured by Sequoia failed to count some ballots marked with gel ink, a problem that was only discovered during the state’s legally mandated hand count of 1% of the ballots cast in the election. The problem, according to Sequoia, was that the machines were calibrated to read only carbon ink — not the dye-based ink found in many gel pens.

Dust, creases, and hardware failures also commonly throw scanners for a loop. In Florida, Sequoia machines sometimes mistook the folds in paper ballots for a vote. Election officials in Volusia County, Florida reported that memory cards in Diebold-manufactured scanners failed during the 2006 election, contributing to a 4.4% error rate. (By 2007, nearly 25,000 Diebold optical scans machines were in use nationwide.) In the District of Columbia during the 2008 election, a defective cartridge caused vote totals to be duplicated into multiple races, and Douglas County, Colorado said it would discontinue the use of ballot scanning machines from Hart InterCivic because of their tendency to read stray marks as votes.

Ballots cast by Black, Hispanic, and young voters are flagged for rejection substantially more often than voters overall, studies show. That’s because they’re statistically more likely to be voting for the first time. According to data analyzed by Dan Smith, a political science professor at the University of Florida, Hispanic voters in Florida during the March 2020 presidential primary were 2.7 times more likely to have their ballots flagged compared with white voters. A California Voter Foundation survey found that during the November 2018 election, the rejection rates for voters in California between the ages of 25 and 34 were nearly double that of all voters.

Barriers to remediation

Manufacturers of optical scanners are often loath to submit to audits, making it difficult to determine the roots of failures. For instance, Diebold conducted a survey of jurisdictions to determine the frequency of its machine’s failures but refused to release results, calling it proprietary information. When the Daytona Beach (Florida) News-Journal contacted the U.S. Election Assistance Commission to compel the release of the information, the commission informed the paper that an official government agency would have to request the action.

Sequoia put up a bigger fight against the District of Columbia Board of Elections after the 2008 debacle. In a response to a request for information about the miscounted ballots, the company said that it found “no anomalies or irregularities in either the data or the internal event logs that can be identified as having caused or contributed to the issue.” Sequoia subsequently issued a report that attributed the problem to human error, ruled out “[e]ndemic hardware and software failures,” and claimed the company had no way to track scanning machine problems. The D.C. Council eventually subpoenaed information about the voting machines’ source code from Sequoia so that it could conduct a more thorough investigation, to which Sequoia objected.

Some scanner issues are likely to fly under the radar because of disparate ballot-counting audit practices. As of mid-2019, only Colorado, New Mexico, the District of Columbia, and Rhode Island require local election officials to perform audits (1) before official results can be certified, (2) that expand to a statewide recount whenever the audit detects serious discrepancies, and (3) that are binding on the official results. An additional 19 states prescribe audits that check a flat percentage (typically between 1% and 3%) of voting machines or precincts, which can detect problems in voting machines but can’t confirm the correct results except in races with large margins. Finally, 28 states finalize official election results without verifying computer-tabulated vote totals, with four states — Florida, Vermont, Virginia, and Wisconsin — allowing audits to be delayed until after winners are certified.

A number of states including California, North Carolina, and Pennsylvania allow voters to track the status of their ballot online but offer little recourse for ballots miscounted by optical scanning devices. A Brennan Center report suggests that prevention is the best remedy to potential issues that might crop up during the election. “Voting system vendors should be required to report both voting system failures and vulnerabilities they have knowledge of,” the report concluded. “Vendors should also be required to make reports to the database when they receive a complaint from a customer (i.e., election official), whether or not they agree that their machine was the cause of the alleged problem; when they receive a warranty claim and/or take some action to satisfy a warranty; when they are notified by a customer of a usability issue that could lead voters or poll workers to operate the system in a way that would lead to disenfranchisement or the recording of an unintended vote; when they conduct an investigation of a reported problem; and when a customer or other person sues them.”

Less than a week before the 2020 election, however, it remains unclear the extent to which vendors will cooperate and whether counties will take steps to ensure historical missteps aren’t repeated.

VentureBeat's mission is to be a digital town square for technical decision-makers to gain knowledge about transformative enterprise technology and transact. Discover our Briefings.