Audit: an independent check of whether a process produced a correct result

An audit is an independent examination of records to test whether a process produced a correct result. An election audit compares reported results with durable records, usually paper ballots, and checks that procedures were followed. All 50 U.S. states and the District of Columbia conduct some type of post-election audit, but the types differ in what they can prove.

Definition

The National Conference of State Legislatures (NCSL) describes a post-election audit as a check “that the equipment and procedures used to count votes during an election worked properly,” often using a hand count of paper records.⁠¹ The U.S. Election Assistance Commission (EAC) defines a post-election tabulation audit as “hand-counting a sample of votes on paper records, then comparing those counts to the corresponding vote totals originally reported,” using the paper records “as the benchmark”.⁠²

The Voluntary Voting System Guidelines 2.0 (VVSG 2.0) define the related term audit trail as “information recorded during election activities to reconstruct steps followed or to later verify actions taken with respect to the voting system,” including “event logs, paper records, error messages, and reports”.⁠³

Origin and history

The modern case for election audits rests on the fact that machines count most votes and their software cannot be assumed correct. The idea of software independence, credited by Lindeman and Stark to Rivest and Wack (2006) and Rivest (2008), made the purpose explicit: errors in software should never be able to change an outcome undetectably.⁠⁴ VVSG 2.0 adopted this as Principle 9, “Auditable,” including the commitment that “the voting system supports efficient audits”.⁠³ National Academies researchers recommended in 2018 that states require a risk-limiting audit “prior to the certification of election results”.⁠⁵

How audits are done

NCSL describes three main types:⁠¹

• Traditional (fixed-percentage) audits “look at a fixed percentage or set number of voting districts or voting machines” and compare paper records with machine results. NCSL reports that 36 states plus the District of Columbia require this type.
• Risk-limiting audits use statistical sampling sized so that a wrong outcome would likely be caught. NCSL counts 7 states requiring them. See Risk-limiting audit.
• Procedural audits check that “the correct process and procedures were followed,” usually including ballot accounting, reconciliation, and “a review of security and chain-of-custody procedures.”

An audit is only as good as its records. Lindeman and Stark write that “the best audit trail is voter-marked paper ballots,” that printed records voters may not inspect “are not as good,” and that “paperless touchscreen voting systems” cannot be audited this way.⁠⁴ They also warn that if a compliance check cannot show that ballots were not altered, added, or lost, an audit “may be mere theater”.⁠⁴ See Chain of custody.

Common confusions and neighboring terms

• Audit vs. recount. A recount re-tallies all ballots, usually triggered by a close margin or a request. An audit samples ballots or checks procedures to test the result.
• Audit vs. risk-limiting audit. An RLA is one type of audit with a defined statistical guarantee; a fixed-percentage audit has none.
• Audit vs. verification. Verification is the general idea of checking; an audit is an organized, independent check with rules set in advance.
• Audit vs. forensic examination. Examinations of equipment by third parties raise their own custody concerns; the EAC advises clear procedures so “the chain of custody is not broken” during third-party work.⁠²

How My Vote Forward uses the term

My Vote Forward uses “audit” in the election sense when it discusses official voting, in Audits and verification, and applies the same logic to its own outputs: every published release carries its method, denominator, version, and corrections so that others can check it against its stated process. The framework does not describe any check of its own demo as an election audit.

Unresolved questions

• Should procedural audit results be published as routinely as tabulation audit results?
• How should audits handle contests too small for a statistical sample to say much?
• What independent check would be appropriate for a civic participation system that publishes only aggregates?