Percentage-Based versus Statistical-Power-Based Vote Tabulation Audits

Abstract

Several pending federal and state electoral-integrity bills specify hand audits of 1% to 10% of all precincts. However, percentage-based audits are usually inefficient, because they require large samples for large jurisdictions, even though the sample needed to achieve good accuracy is much more affected by the closeness of the contest than population size. Percentage-based audits can also be ineffective, since close contests may require auditing a large fraction of the total to provide confidence in the outcome. We present a plausible statistical frame-work that we have used in advising state and local election officials and legislators. In recent federal elections, this audit model would have required approximately the same effort and resources as the less effective percentage-based audits now being considered.