Voting Blogs: Ballot-level comparison audits: central-count | Andrew Appel/Freedom to Tinker
All voting machines these days are computers, and any voting machine that is a computer can be hacked to cheat. The widely accepted solution is to use voting machines to count paper ballots, and do Risk-Limiting Audits: random-sample inspections of those paper ballots to ensure (with a guaranteed level of assurance) that the election outcome claimed by the computers is the same as you’d get by an accurate count of the votes actually marked on the paper ballots. An RLA is any statistical/logistical method that guarantees a quantifiable risk limit. Most RLA methods are much more efficient than a full recount, but some RLA methods are more efficient than others, especially in close elections: the closer the margin of victory, the more ballots you have to random-sample to guarantee the risk limit. One simple method, the Ballot Polling Audit, randomly samples individual ballots from among all the batches of ballots in the entire election, and (by human inspection of the paper ballot) counts how many are for each candidate. If you sample enough ballots, and that “poll” comes out the same way as the outcome claimed by the voting machines, then you get real assurance. (If the “poll” doesn’t come out the same way, you can do a bigger poll or a full hand recount.)