Audit types explained
A tabulation audit checks the machine count against the paper ballots. States used four broad kinds of audit in November 2024.
Fixed-rate hand counts
People hand count the ballots from a set share of precincts, machines or batches (typically 1% to 5%) and compare the vote totals with the machine count, which may or may not be a total that appears in the canvass. Ballots are viewed by people but the ballots are not compared with corresponding cast vote records so it is unlikely that any individual ballot interpretation problems would be discovered. Many states expand the count if the count difference passes a threshold. The sample is the same size whether the race was a landslide or a near tie. This is the most common kind of audit.
For example: California (1% of precincts), Arizona, New York, Minnesota.
Tiered audits
Also a hand count of a share of precincts, but the share grows as the margin shrinks, so close races get a larger sample. The tiers are set in law, not calculated from statistics.
For example: Oregon (3% of precincts, rising to 5% and 10% in closer races), Kansas (an extra 10% for races within 1%), New Mexico.
Risk-limiting audits (RLAs)
The sample size is calculated from the contest margin and a chosen “risk limit”: the largest acceptable chance that a wrong outcome goes uncaught. Ballots are drawn at random, one at a time or in batches, and the audit continues until there is strong evidence that the reported winner really won, or until every ballot has been counted by hand. A comparison RLA checks each sampled ballot or batch against the machine’s record of it; a ballot-polling RLA looks only at the paper ballots. An RLA confirms the outcome of the races it audits; it does not check every vote total. RLAs generally audit a small number of contests: each one needs its own sample, a close contest can need a large one, and races that cross county lines need the state to coordinate the sampling.
For example: Colorado, Georgia, Rhode Island, Virginia, Washington.
Ballot image audits (BIAs)
Separate software, independent of the voting system, re-counts scanned images of the ballots and compares its totals with the official ones. The images are either the voting system’s own or come from a fresh scan of the paper on separate equipment; only a fresh scan can catch a ballot the voting system’s scanner imaged badly. It can cover every ballot and every race quickly, but it checks images of the ballots, not the paper itself, and no person reads a ballot.
For example: Maryland and South Carolina (every ballot statewide), and Florida, where 28 counties chose an independent re-scan of at least 20% of precincts. Vermont and New Hampshire re-scan the ballots of a few towns or machines on separate equipment.