In Count Every Vote, I expand on this last point: Because repeated scientific studies show that software-driven voting systems provide us with no rational basis to trust reported results, in those jurisdictions that use computerized voting systems and that do audit the voter-completed ballot, only a tiny percentage of all ballots cast is reliably counted. This represents well over 90% of US elections with less than 5% of all ballots reliably counted.
Unless the audit is done at the precinct, on election night, before observers of different political faith, Novick points out that even the audit count is not reliable, since continuous observation of the ballots ends after election night. Once this chain of custody is broken, there is no proof that the ballots counted during the audit "represent the actual ballots cast at election."
New York's 2005 electoral scheme (ERMA) seeks to mimic this unsound and unconstitutional practice of only counting a small percentage of all ballots cast, allowing the audit to be performed two weeks after the election, when continuous observation of the ballots has long ended.
New York has a choice right now. It can stand its ground, as it has for the past six years, and save its sound, secure, functioning electoral system, or it can follow the rest of the nation over the cliff by implementing non-securable, unreliable, computerized voting technology that removes all rational basis for confidence in reported results.
The ship has not yet sailed, New York: save the levers.
As an important aside, centralizing the voter database needlessly exposes NY's 12 million registered voters to disenfranchisement via politically targeted "caging" and identity theft, a rising phenomenon directly related to the use of non-securable computerized databases. We must question any law that requires a statewide computerized voter database, given that the electronic databases of the Pentagon nor DHS are immune from attack. Decentralized databases protects against statewide fraud, limiting would-be hackers to county-level vs. state-level data. Likewise, defrauding a lever machine affects only that machine, whereas hacking a software driven machine can infect the entire county.
3 July 2005, submission to the Houston EAC Hearing by Rady Ananda. Chart of election incidents reported (EIR) in Ohio's 2004 election, by county and by vendor. http://www.electionassessment.org/Submissions/2005-06-29EAH/Ananda_R/index.html An updated version was published in What Happened in Ohio: A documentary record of theft and fraud in the 2004 election, by Robert J. Fitrakis, Steven Rosenfeld and Harvey Wasserman, NY: New Press, 2006, 116-123. Note how often vote switching occurred in the EIRs column. Optical scanners are just as vulnerable as touch screens to this type of software attack.