RLAs

Risk-limiting audits (RLAs) are post-election procedures that provide statistical evidence that a reported election outcome is correct, with a pre-specified maximum probability that an incorrect outcome is certified, known as the risk limit. When implemented, RLAs can confirm results with far fewer ballots than a full hand recount, while preserving transparency and verifiability. Originating from statistical audit theory and election integrity initiatives, RLAs have been developed to handle various voting systems and ballot formats.

There are two primary families of RLAs. Ballot-level comparison audits compare each scanned ballot or ballot image to the corresponding cast vote record and tally, using deviations to bound the probability of an incorrect outcome. Ballot-polling audits select ballots at random and use the sampled results to infer whether the reported outcome is correct, typically via sequential statistical testing. Hybrid approaches combine elements of both. The auditing process is driven by a chosen risk limit, which dictates the required strength of evidence to stop the audit.

Core steps include: selecting a target risk limit; defining the scope and hypotheses; choosing a sampling plan; tallying the sampled ballots; computing a statistical metric; and deciding whether to stop or to continue, potentially escalating to a full hand tally if evidence remains insufficient.

Advantages of RLAs include formal guarantees of correctness, efficiency relative to full recounts, and improved public trust. Limitations involve dependence on accurate ballot records or ballot images, robust sampling procedures, and sufficient administrative capacity to administer the audit. RLAs have been studied extensively and implemented in several jurisdictions around the world, contributing to the broader movement toward verifiable elections.