Colorado -- a crucial swing state -- completed its $13 million database this year after firing Accenture in 2005. A little-known Oregon company named Saber, which has created databases for 11 states, replaced it. Accenture retained its contract in Pennsylvania, though problems occurred there as well. In 2005, one state official called the $20 million system "seriously if not fatally flawed."
HAVA requires databases to have "adequate technological security" but doesn't specify details, such as encryption. And although the databases interface with every county election office, access controls haven't been developed in some states.
A 2006 audit of Florida's registration system found that the state hadn't established adequate access levels for various users and had no process for maintaining or monitoring audit logs, making records vulnerable to theft and manipulation. A June 2008 follow-up found some of the same problems. One former election office employee, for example, still had access to the database three months after leaving his job.
In 2006 in Denver, electronic poll books made by Sequoia Voting Systems crashed extensively, causing long lines that resulted in an estimated 20,000 voters leaving polls without voting. During Georgia's primary this year problems with e-poll books made by Diebold Election Systems led to voting delays up to three hours long.
Despite various issues, Kay Stimson, spokeswoman for the National Association of Secretaries of State, says the registration databases are ready, and states are confident they'll perform well for the election. She acknowledges, however, that issues over HAVA matches are still a concern.
"Generally speaking, the uncertainty that hangs over the process, including uncertainty that results from election challenges and litigation introduced shortly before Election Day, creates a greater likelihood for problems or confusion at the polls," she said.
HAVA leaves it to states to decide how to conduct matches. Some states require an exact match with the Social Security Administration database and only a substantial match with motor records. Others require an exact match for a voter's Social Security number, first and last name, and month and year of birth.
Exact matching, however, could mean that a woman who recently married and changed her name would fail to match government records containing her maiden name. Voters who have double last names or unusually spelled names might also fail. Everything depends on how a state's matching algorithm is designed.
Last month Wisconsin, whose database just became operational, conducted a test of 20,000 voter names against motor vehicle records and found 20 percent with mismatches, due mainly to typos and transposed numbers. Among those who failed to match were four members of the state's Government Accountability Board (.pdf), which conducted the test. Thomas Cane, the board's chairman and a retired judge, failed because he was listed by his full name, R. Thomas Cane, in his driver's record.
A recent report from the Academies of Sciences noted that "many (if not most) of the matching procedures used by the states have been developed on the basis of intuitive reasoning without further systematic validation or mathematically rigorous analysis, do not reflect the state of the art in matching techniques, and have not been validated in the market, scientifically, or otherwise."