|
Abstract:
|
Post-election audits to verify electronic vote counts have received growing attention. Previous research focused on shortcomings of fixed, or tiered, percentage audits and on the development of a method to determine the audit sample size based on power. A power-based method ensures a high probability of detecting a miscounted precinct if enough exist to have altered the election outcome. If an audit reveals no miscounted precincts it is complete. The procedure to follow, however, when miscounts are found remains unclear. The detection of miscounts may not require further auditing to verify the winner. We propose a trigger for additional auditing that is based on a modified Hoeffding bound. This trigger is compared, through simulations, to one based on a bootstrap confidence interval estimate of the net gain in votes that the reported loser would obtain should a 100% audit be conducted.
|
