Risk-Limiting Audits

A risk-limiting audit (RLA) of the tabulation of an election contest checks some voted paper ballots or voter-verifiable paper records in search of strong evidence that the reported contest outcome was correct – if it was.  A risk-limiting audit can stop as soon as it finds strong evidence that the reported outcome was correct. Or, if the reported outcome was wrong because ballots were miscounted in the tabulation, an RLA is very likely to lead to a full recount that corrects the outcome.

RLAs avoid checking ballots unnecessarily. Unlike fixed-percentage audits, risk-limiting audits base the number of ballots selected for audit on the specifics of the contest. Contests with a wide margin can be audited with very few ballots, freeing up resources for auditing closer contests. Closer elections generally entail checking more ballots. Even in close contests, risk-limiting audits can often provide confidence in correct outcomes with a modest amount of effort.

More formally, a risk-limiting audit with a 5% risk limit has at least a 95% chance, if an outcome is wrong, of leading to a full hand count that will correct it. (The risk limit is the corresponding 5% maximum chance that the audit will not lead to a full hand count if the outcome is incorrect.) The chance of correcting a particular wrong outcome may be considerably larger than the guaranteed minimum, depending on the actual vote counts and other circumstances.

A common misunderstanding is that “risk” has to do with the chance that an outcome is wrong, or the percentage of outcomes that are wrong. Actually, risk quantifies how well the audit performs when outcomes are wrong. An analogy may help: Think of wrongness as a medical condition that some outcomes have. Then a risk-limiting audit with a 5% risk limit has a 95% guaranteed minimum chance of both diagnosing and curing wrongness, if it exists – no matter how rare or common wrongness is.

RLAs are statistically sound. The American Statistical Association endorses and recommends risk-limiting audits.

Download PDF Download this page in PDF format