Primary elections in Ohio last May provided an object lesson in what can go wrong with an election — even when high-tech voting machines produce paper backup records.
In Cuyahoga County (Cleveland), nearly 10 percent of the rolls of paper on which results were printed out were “destroyed, blank, illegible, missing, taped together or otherwise compromised,” the San Francisco-based Election Science Institute (ESI) reported to the County Commission last August in a 234-page report.
“Discrepancies between paper record and the electronic record were . . . pervasive,” ESI said. “One likely result is diminished public confidence in a close election.”
Paradoxically, the election system that ESI examined had been upgraded in recent years precisely to assure reliability. Ohio is one of 25 states that require a paper record of each vote, even when voters use direct recording electronic (DRE) machines, such as those that Cleveland voters used. These touch-screen devices weren't designed originally to need paper.
But computer scientists, activists and some politicians saw a flaw in that design.
Without a paper ballot, “You cannot assume that what you put in is what you get out,” says Richard A. Celeste, a former governor of Ohio who co-chaired a study of voting-machine technology for the National Research Council of the National Academies. “You need ways to monitor and audit results.”
The drive for paper backup follows a nationwide switch to electronic voting machines. In the Help America Vote Act of 2002 (HAVA), Congress authorized handing out $325 million to states so they could replace old-fashioned punch-card and lever-action machines. Lawmakers were responding to the presidential election crisis of 2000, when Florida vote-counters tried to determine who won by examining paper ballots for “hanging chads.”
Many states and local governments turned to optical-scan machines, which “read” voters' markings on a paper ballots and count the results. By its nature that system comes with its own paper backup.
In congressional elections this November, about 40 percent of U.S. voters will be making their choices on optical-scan devices, and a roughly equal share of voters will be using DREs.
Michael Shamos, a computer scientist and lawyer at Carnegie-Mellon University in Pittsburgh, Pa., says paper records aren't necessarily more reliable than computer memories. “If the machine cannot be trusted — which is the working hypothesis of paper-trail proponents — then it cannot be trusted to deal with the paper trail safely,” he writes.
Nonetheless, says Aviel D. Rubin, a computer-science professor at Johns Hopkins University, if election results are challenged in a precinct that uses paperless computers, “There's nothing to count.” Rubin and others have been pressing that argument even as DRE electronic voting machine sales have boomed. Revenues of the four industry leaders and three smaller firms total from $350 million to $550 million a year, estimates Michael Kerr, director of the International Technology Association of America's Election Technology Council, the trade group for voting-machine manufacturers.
DRE advocates say the machines all but eliminate the problems of unreadable or incorrectly marked ballots, known in the election trade as residual votes and undervotes. “DREs provide dramatic improvements” over mechanical systems, Kerr says. As for paper-printout requirements, “I don't think we have any particular opinion.”
Other DRE defenders are quick to point out that the machines are not known to have altered the results of any elections. “There were no substantiated reports from any state of compromised elections due to security flaws that involved computer hacking or similar attacks in 2004,” the Congressional Research Service reported, though acknowledging that some computer fraud can't be spotted.
However, a Government Accountability Office report last year found that some electronic voting machines did not encrypt votes or audit logs, making it possible to alter the files “so that the votes for one candidate could be recorded for a different candidate” without being detected.
HAVA does require an elementary level of hard-copy backup — a requirement that every voting machine produce a “permanent paper record with manual audit capacity.” But Rubin and other DRE critics say the gold standard is a “voter-verified paper audit trail (VVPAT).”
This system was the brainchild of Rebecca Mercuri, president of Notable Software, a Hamilton Township, N.J., consulting and forensic-investigation firm. In a VVPAT system, the voters don't “cast” their vote until they have seen a paper printout of the choices made on the touch screen. If the printout is wrong, that printout is voided, and the voting process begins again. “By that casting action, the ballot is verified,” Mercuri says. “Now it is the ballot of record.”
Rep. Rush Holt, D-N.J., has proposed a bill to require VVPAT systems nationwide, which is awaiting House Administration Committee action.
Mercuri supports Holt's bill. But she and many other VVPAT advocates make no secret of their view that the paper trail is merely a stopgap. They argue that computer memories are too vulnerable to tampering and breakdown to be entrusted with elections. “Abolish DREs entirely,” urges journalist Ronnie Dugger of Cambridge, Mass., who wrote the first account of potential problems with computerized voting in 1988. “There should be no vote counting by secret and invisible ballots.”
As Cleveland's troubles in its primary elections in May would seem to indicate, even VVPATs don't offer a completely reliable backstop. Mercuri, Rubin and others say optical-scan systems offer far more protection against technological problems and high-tech fraud.
But the manufacturer of Cleveland's DRE machines, Diebold Election Systems Inc. of Allen, Texas, didn't take the ESI report lying down. The company blamed problems on poll workers and accused ESI of shoddy work. “There would seem to be apparent weakness in the skills of ESI or the investigatory methodology employed by them,” Michael E. Lindroos, the company's vice president, wrote to county commissioners on Aug. 16 ESI countered that technical data it had requested was still not forthcoming and that flaws and questions identified during the investigation hadn't been resolved.
None of the give-and-take satisfied William Ritter, a high-school history teacher who had run for the Democratic nomination for the state House of Representatives. Between the first unofficial count and the official count two weeks later, Ritter went from winner by 115 votes to loser by 178.
Given all the vote discrepancies, Ritter asks, “How can they tell me I lost?”