Two important components of an audit procedure are the sample size and the decision rule for expanding the audit. This paper describes a method for determining the audit size that ensures a high probability of detecting miscounts if election-altering ones exist. Then, two possible “triggers” for further auditing are developed: a modified bootstrap confidence interval and a modified Hoeffding bound. Both estimate the net gain in votes for the originally reported loser if a full audit were conducted. In simulations, both methods maintained high power for various miscount situations, with the Hoeffding bound having slightly lower false positive rates.
Lim, Katherine, "Post-election audits: statistical power based methods and a trigger for further auditing" (2008). Honors Projects. Paper 9.
© Copyright is owned by author of this document