A Review of Robust Post-Election Audits

In August of 2018, the Michigan Bureau of Elections and the city clerks of Kalamazoo, Lansing, and Rochester Hills partnered with the Brennan Center for Justice, Professor Ron Rivest and Mayuri Sridhar of MIT, Dr. Philip Stark and Kellie Ottoboni from the University of California, Berkeley, Jerome Lovato of the U.S. Election Assistance Commission, Verified Voting Foundation, and the Voting System Technical Oversight Program at Ball State University (the “RLA Team”) to conduct the first risk-limiting audit (“RLA”) pilot in Michigan.

While the main goal was to provide Michigan election officials with a hands-on learning experience about RLAs, the “gold standard” of post-election tabulation audits, participants gained broadly-applicable insights about best practices. We hope this report can serve as a resource to election officials across the country who are considering adding RLAs to their election security procedures or conducting similar pilots. We have designed it to be useful for election officials with varying levels of familiarity with RLAs.

The executive summary and the project overview provide a high-level overview of the project, introduce the different audit methods used, and list the voting systems and RLA tools used in each jurisdiction. A risk-limiting audit definition and a chart that summarizes the pilot audit results in each jurisdiction are also provided.

Next, election officials offer their insights on the process, the pilot and their goals moving forward. State election officials provide the background to this unique partnership and add their perspective on the project. Importantly, they discuss lessons learned through piloting different audit methods and random ballot selection methods, and how they plan to move forward in the future.

Local election officials then provide a detailed overview of their individual experiences during the pilot. Each discusses their workload, communication strategy, implementation strategy and lessons learned. They also provide recommendations to state election officials.

Finally, two members of the RLA Team describe the foundational mathematics, specific procedures, sampling methods and RLA software tools.

If you have additional questions about the procedure, the pilot, or any other aspect of this partnership, as we hope you do, please do not hesitate to contact anyone involved.

State and local election officials serve as our democracy’s last line of defense against malevolent foreign actors, equipment malfunctions, and human errors which may impact election results. Unfortunately, this responsibility often comes with few resources and heightened public scrutiny. One smart and effective tool available to election officials facing this reality is the post-election risk-limiting audit (“RLA”).

This special type of audit uses statistical methods and a manual review of paper ballots to check the accuracy of reported election outcomes. Specifically, RLAs are designed to provide assurance that the reported winner did in fact win the election, or in the alternative, to correct errors caused by cyberattacks, bugs, misconfiguration, or human error, if any combination of those altered the reported outcome. While the underlying math may be challenging for non-mathematicians to understand, the procedures to conduct such audits were shown to be straightforward.

In August of 2018, the Bureau of Elections agreed to partner with the RLA Team to conduct the first risk-limiting audit pilots in Michigan. The RLA Team worked directly with election officials to understand the relevant election administrative procedures and practices. They used this information to draft audit instructions (called audit protocols) for each locality. Each participating municipality had one day to conduct their pilot between December 3–5, 2018.

This approach produced many practical lessons. Most importantly, Michiganders gained confirmation that RLAs are possible in their state, which relies primarily on precinct-based voting on election day. Many other lessons related to the procedure, voting systems, and messaging were drawn from this groundbreaking collaboration. These lessons include:

• Risk-limiting audits are an effective tool that can be implemented in Michigan, by Michigan election officials, using Michigan-certified voting systems.
• Risk-limiting audits are not procedurally difficult to implement by election officials.
• Risk-limiting audits can be implemented with minimal changes to pre-election and Election Day administrative procedures.

Overall, Michigan election officials were impressed with the results from the pilots, especially the potential to greatly improve post-election audit efficiency. While further work and additional pilots are necessary, this pilot equipped election officials with practical information necessary to make important election security policy decisions in Michigan. They currently plan to take the pilot to the next step by conducting RLAs at the county level after the May 2019 elections.