Civil rights groups filed a lawsuit against the Pennsylvania Department of State last week calling for paper ballots to be provided in any precinct where half the voting machines fail on Election Day. The complaint asserts that 10 percent to 20 percent of the direct-recording electronic voting machines used in Pennsylvania are likely to fail on Nov. 4. The machines will be used in 34 percent of counties around the United States. Are they really so bad that they fail one out of five times?
No. The 10 percent to 20 percent figure cited in the Pennsylvania lawsuit is somewhat misleading. The low estimate comes from an opinion piece published by the National Academy of Engineering last year. According to the author, Michael Ian Shamos, "it has been reported anecdotally" that 10 percent of the machines fail "in some respect" on Election Day. That means the machine required some technical intervention, not that it was necessarily taken out of service. The high estimate of 20 percent comes from a 2005 study conducted by the state of California. * In that paper, the authors were testing only one of the six machines used in Pennsylvania—the Diebold (now Premier) AccuVote TSX, a machine whose software has since been updated. Furthermore, 14 of the 34 documented failures were printer jams, which won't be a problem in Pennsylvania since the state doesn't produce paper voting receipts.
The fact is that no one really knows how often electronic voting machines fail. The Election Assistance Commission—an independent governmental agency charged with establishing election standards—doesn't collect comprehensive statistics on failure rates. (Various nonprofits, such as Election Protection and VotersUnite!, do collect individual complaints.) However, according to federal standards set in 2002, machines may fail as often as once in 163 hours and still make certification. If the chance of a failure were randomly distributed throughout that 163-hour period, a given machine would have up to around an 8 percent chance of breaking down during regular use on Election Day. But these standards define "failure" quite broadly—a software glitch that causes the machine to freeze up for 10 or more seconds, for example, would count. (Individual states don't have to follow the federal guidelines, though many of them do.)
Got a question about today's news? Ask the Explainer.
Explainer thanks Matt Bishop of the University of California-Davis, John Gideon of VotersUnite!, Joseph Lorenzo Hall of University of California-Berkeley and Princeton University, Lawrence Norden of the Brennan Center at New York University, Michael Shamos of Carnegie Mellon, and David Wagner of University of California-Berkeley.
Correction, Oct. 28, 2008: The original sentence stated that the 2005 AccuVote TSx study was part of a top-to-bottom review of California's voting systems. It was conducted under the state's regular certification process; the top-to-bottom review took place in 2007. (Return to the corrected sentence.)