In early December, I went to Michigan to facilitate pilot post-election risk-limiting audits in three cities. This was the first time that Stratified Union-Intersection Tests of Elections (SUITE), a new method for "hybrid" risk-limiting audits that I helped develop, has been used in practice.
A risk-limiting audit (RLA) is a statistical check that the reported outcome of an election is correct. The procedure limits the chance that an incorrect outcome will go undetected, if the reported outcome is in fact wrong. RLAs require a trustworthy paper trail and involve hand examining a random sample of paper ballots.
SUITE combines two types of RLAs: ballot polling, which involves recording the number of votes for each candidate, and ballot comparison, which involves comparing paper ballots to their electronic record and counting the number of discrepancies. Ballot-level comparison audits require looking at fewer ballots than ballot polling, but can only be done when ballots can be matched to their electronic records. In many cases, only absentee ballots can be matched to their electronic records. SUITE allows you to use these two strategies side-by-side for a single RLA of absentee and in-person ballots.
I wrote a Jupyter notebook tool to do the SUITE risk calculations, to determine the necessary sample size, and to sample the ballots using a cryptographically secure pseudo-random number generator (see Related Articles, below).
Rochester Hills, Lansing, and Kalamazoo participated in the pilots of their November 2018 election. I traveled to each city with a team of election auditing experts from MIT, the NYU Brennan Center for Justice, the Electoral Assistance Commission, and Democracy Works. Each day began with presentations on RLAs and a dice-rolling ceremony that used 10-sided dice select the 20-digit seed for the pseudo-random number generator used to sample ballots.
Teams of two retrieved the sampled ballots from ballot bags and recorded the votes cast on them. Even in Lansing, where the contest had a 10% vote margin and required sampling 260 ballots, the whole process only took several hours. Compared to doing full manual recounts, RLAs can be a more efficient way to give assurance that the correct candidate was deemed the winner.
Philip Stark first developed RLAs in 2008. Since then, they have been piloted in California, Colorado, Indiana, Virginia, Ohio, and Denmark. Colorado started conducting routine statewide RLAs in 2017, and Rhode Island passed a law in 2017 requiring routine statewide RLAs starting in 2020.
Risk-Limiting Audits by Stratified Union-Intersection Tests of Elections (SUITE)
September 6, 2018 | Kellie Ottoboni, Philip B. Stark, Mark Lindeman, Neal McBurnett | International Joint Conference on Electronic Voting
Simple Random Sampling: Not So Simple
February 3, 2017 | Kellie Ottoboni | BIDS Blog: Data Science Insights
Random Sampling: Practice Makes Imperfect
October 26, 2018 | Philip B. Stark, Kellie Ottoboni | arXiv.org