DiffSum - A Simple Post-Election Risk-Limiting Audit

TL;DR

DiffSum is a straightforward post-election audit method that sequentially samples ballots until a statistical condition confirms the reported winner, offering a simple and effective risk-limiting audit for two-candidate elections.

Contribution

This paper introduces DiffSum, a new simple sequential ballot-polling audit method with error bounds, improving ease of implementation over existing techniques.

Findings

DiffSum effectively controls error rates through simulations.

Compared to Bravo, DiffSum offers a simpler implementation.

DiffSum provides reliable risk-limiting properties for two-candidate elections.

Abstract

This paper presents DiffSum, a simple post-election risk-limiting ballot-polling audit for two-candidate plurality elections. DiffSum sequentially draws ballots (without replacement) until the numbers $a$, $b$, of votes for candidates $A$, $B$ satisfies $a>b$ and $(a-b)^2 > c(a+b)$, where $A$ is the reported winner and $c$ is a suitably chosen constant. Bounds on the error rate (chance of approving an incorrect election outcome) are obtained via simulations. The method is compared with the Bravo method of Lindeman, Stark, and Yates.