Improving Risk-Limiting Election Audits: A Blockchain-Based Solution
by Ann Duke and Caroline Morin
In late 2020, leading into January 2021, scores of American citizens initiated multiple violent protests in Washington DC, united behind the slogan, “stop the steal,” referring to President-Elect Joe Biden’s win of the November 2020 presidential election.1 While the intensity of this dissent exceeds recent memory, fear of fraudulent election results has occurred at least twice in American history, due to the alleged Chicago Mayoral election manipulation by Mayor Daley in 1960, and the hanging chads issue of the 2000 presidential election.2
In this era of renewed distrust in democratic elections, blockchain — a cryptographically secure, distributed ledger system — may be evidenced as a solution.3 A blockchain-based audit system allows for increased security, transparency, efficiency, and traceability of assets; these are four key features that can quell fears of technological errors related to the incorrect tabulation of election margins.4 Fear differs from distrust because it can be solved by attacking disinformation — its root cause. Fear raises adrenaline and creates an emotional response to a perceived threat, the threat here being the loss of a cornerstone of American governance: free and fair elections.
A democracy’s power comes from its citizens, and if its citizens start to believe in a perceived threat to their democracy, it arguably ceases to exist. Risk-limiting audits (RLAs) provide a check on the American democratic process, which allows the public to hold election officials accountable and see the facts behind their perceived threat. It is important to acknowledge that RLAs are separate from voting and vote tabulation; they do not change election results, only help to verify their accuracy.
RLAs check the accuracy of election results by using random sampling to verify the correct tabulation of elections.5 To conduct an RLA, auditors randomly select ballots and then count them by hand until they meet the risk-limit, at which point there is sufficient evidence that the election was correctly counted.6 The number of votes counted increases as the electoral margin-of-victory decreases. If the election results were correct, then the findings of the audit would match the election results. If an issue existed with the election, RLAs would have a statistically significant chance of finding it because RLAs serve as a final check on accuracy before certification of the election.7
RLAs, considered to be the gold standard of testing election accuracy, face significant barriers to implementation throughout the US.8 RLAs require significant labor and capital, effectively barring their use in forty-six states, which do not have statutory requirements to conduct RLAs after each election. Using blockchain to tabulate the votes instead of human auditors can considerably reduce how much labor and capital an RLA needs while improving the audit’s accuracy. Reducing barriers to RLA implementation would likely lead to greater proliferation and could residually alleviate the fear of elections being incorrectly counted.
Using blockchain technology instead of human auditors would reduce barriers and increase transparency in the electoral process. Blockchain’s distributed ledger technology allows the public to conduct RLAs on their own. Elevating the public’s role from watcher to active participant allows for a reduction in fear and a greater sense of ownership over the American democratic process.
Reducing Barriers to Risk-Limiting Audits
The high amount of labor and capital required for each RLA serves as a significant barrier to implementation.9 In a longitudinal analysis in which studied elections had both tight and wide margins, the presence of RLAs resulted in an overall reduction in the need for full recounts.10 While in the long term RLAs reduce costs, in the short term, given high rates of election accuracy, justifying the high price of implementation could prove difficult. However, reducing the associated costs could change this calculation.
The incorporation of manually counting ballots increases costs associated with RLAs. To avoid repeating underlying errors, audits must utilize different election equipment than the machinery that initially tabulated ballots.11 To combat this issue, and ensure that human intent is accounted for, most audits employ a hand-tabulation method; however, this is “time-consuming,
labor-intensive, and prone to human error.”12 While manual audits require no additional purchase of technology, they carry steep costs: a November 2017 RLA in Colorado cost about 500 USD for 516 ballots.13 Applying this cost approximation to the RLA of the November 2020 presidential election in the US state of Georgia, which audited every ballot due to the tight margin, would have amounted to roughly 5 million USD.14
Using blockchain to count votes during the RLA process can reduce barriers to implementation by minimizing labor costs.15 A 2019 study found that the tabulation of ballots can feasibly occur on blockchain in a way that fulfills democratic obligations.16 In contrast to the high labor costs associated with a hand-count, conducting tabulation on blockchain minimizes the involvement of people and, by effect, the cost of labor.17 Blockchain’s decentralized storage and cryptographic hashing makes the data storage immutable, meaning that, after data is uploaded to the blockchain, it cannot be manipulated.18
Blockchain has the potential to transform the RLA tabulation methodology. However, blockchain cannot eliminate every barrier to implementing an RLA. Blockchain, as an emerging technology, has its limitations. For instance, 70 percent of municipalities used paper voting technology in the November 2020 election, and blockchain lacks the capacity to audit these ballots.19 Interpreting human intent requires manual auditing.20 For ballots filled out by a machine, blockchain tabulation eliminates the necessity of human auditors and the costs associated, an act that can reduce some barriers for states to implement RLAs.
Former US President Donald Trump propagated unfounded questions surrounding the accuracy and dependability of Dominion Voting, a voting technology manufacturer, after the November 2020 presidential election. According to Trump’s social media posts, the software caused “tens of thousands of votes [to be] stolen … and given to Biden.”21 Further conspiracy theories, spread by supporters and posts by the former president himself, implied that Dominion deleted votes from the software and switched votes from Trump to Biden.22 These claims, while grounded in fiction and incorrect knowledge about vote tabulation, had a real impact. Dominion’s software was listed as one of the grounds for overturning the election, in an Amicus Curiae brief written by the Attorney General of Texas and agreed to by seventeen other states.23
Increased implementation of RLAs would increase transparency and decrease unfounded fears of mistabulation and vote-changing. Before an RLA occurs, when voters cast their ballots, most states require voter verifiability: before a voter casts their ballot, they must be able to observe that their ballot was marked the way they intended.24 Dominion Voting and other voting software companies follow this requirement and create machines that meet the voter verifiability standard and create a paper record of it; therefore, even if deleting or switching votes did occur, a voter-verified paper trail would ensure that the election still produces the correct winner.25 This paper record of votes is saved, and during an RLA, each paper ballot has an equal chance of random selection to test its accuracy.26 If a problem with the software which was big enough to change the margins of the election existed, RLAs would have a strong chance of finding the error.27
Audits have enormous power to reduce fears that citizens have of unfairly counted elections; thus, many states take measures so the public can see the audit conducted accurately and honestly. Almost all states that implement RLAs conduct their audits in the view of the public and publish the results.28 However, an audit conducted via blockchain could allow states to expand this transparency by letting voters conduct RLAs on their own. This takes the voter-verified ballots a step further, as voters would not only be able to confirm that their ballot was properly filled out, but conduct an audit of the ballots themselves. Previous research in related fields has shown that allowing people to take ownership of a solution can boost their confidence in it.29
In September 2019, Utah County, UT proved that the public can conduct audits on their own with only a laptop and an internet connection. However, the county also used blockchain to cast the ballots, a method proven to have several large security issues.30 Still, the program successfully increased transparency by allowing people to conduct their own audits after receiving the same data as the elections board, a how-to guide, and an instructional video.31 The presence of this blockchain-based audit is indicative of officials’ potential willingness to allow the public greater ownership over RLAs, building on the ownership associated with voter-verified paper ballots.
While blockchain can enhance RLAs, the whole voting process cannot move online. Many studies have found blockchain-based vote casting, specifically with the Voatz app employed in Utah County, has large security issues, including the potential alteration of votes, undetectable errors, and general incompatibility with the voting process.32 These limitations for blockchain-based voting do not carry over to applying blockchain in an RLA, however. In general, the US intelligence and defense communities, along with foreign governments, have found minimal security issues in using blockchain in an RLA, leading to the incorporation of blockchain into other aspects of government.33 It should be noted that while the blockchain contains several cryptographic tools, they do not prevent hackers from infiltrating the chain. They do, however, make the attempted manipulation obvious.34 The incorporation of blockchain into RLAs can consequently boost their power to detect errors in the process by virtue of its ability to recognize cyber-attacks and foreign probes.
Bridging the Digital Divide
Previous literature written posits that e-government acceptance is a function of trust in the internet, trust in the government, and perceived risk. When this model is applied to using blockchain to perform RLAs, it does not hold because of distrust in the government and any government-run solutions, as evidenced by the “stop the steal” riots.35 However, blockchain’s design circumvents the need for both trust and a central institution. Therefore, the Technology Acceptance Model (TAM), which asserts that acceptance of technology is a function of perceived usefulness and perceived ease of use, more accurately represents the situation.36 Through the application of TAM, if the public perceives and comprehends the usefulness of blockchain in RLAs, regardless of their trust of the government, the model predicts acceptance.
In the twenty-first century, the proliferation of technology has empowered citizens to promote democratic principles and hold election officials accountable. Technology has admittedly also allowed conspiracy theories, such as the “stop the steal” movement, that may have once remained at the margins of society, to become mainstream. This context of increased volatility and individual shaping of democracy requires more transparent governance than ever before.
When election officials have an opportunity to increase transparency and, by effect, public trust, their oath of office obligates them to take it. Integrating blockchain into RLAs provides several opportunities to enhance the process, including greater transparency and reducing barriers to implementation. While increasing the prevalence of RLAs will not solve the rampant distrust that voters may carry, it is a start. Verification that votes are counted accurately and fairly is the best way to actually “stop the steal.”
1 Davies, Emily, Rachel Weiner, Clarence Williams, Marissa J. Lang, and Jessica Contrera. “Multiple People Stabbed after Thousands Gather for Pro-Trump Demonstrations in Washington.” Washington Post, December 12, 2020. https://www.washingtonpost.com/local/trump-dc-rally-maga/2020/12/11/8b5af818-3bdb-11eb-bc68-96af0daae728_story.html.
2 Speel, Robert. “Four Times the Results of a Presidential Election Were Contested.” Smithsonian Magazine, November 4, 2020. https://www.smithsonianmag.com/history/rigged-vote-four-us-presidential-elections-contested-results-180961033/.
3 Blockchain is an efficient data structure because all information is stored in a chain that each person who uses the blockchain has access to; thus, the information is not stored by one central actor but rather by many actors. This ensures that if a malicious actor does attempt to manipulate the blockchain, their actions are obvious to all of the actors that steward the ledger.
Nelson, Paul. Primer on Blockchain. United States Agency for International Development, n.d. https://www.usaid.gov/sites/default/files/documents/15396/USAID-Primer-Blockchain.pdf.
4 It is important to emphasize that this paper is discussing the introduction of blockchain in the vote tabulation process, it is not advocating for the casting of votes on blockchain, which has serious security concerns. These concerns will be addressed later in the paper.
5 Goodman, Susannah, Philip Stark, and Mark Lindeman. Risk-limiting Audits: Frequently Asked Questions and Answers. Verified Voting, Brennan Center for Justice, and Common Cause, n.d. https://verifiedvoting.org/wp-content/uploads/2020/05/RLAs-FAQ.pdf.
6 “Risk-limiting Audits.” National Conference of State Legislatures. Last modified February 17, 2020. Accessed January 10, 2021. https://www.ncsl.org/research/elections-and-campaigns/risk-limiting-audits.aspx.
7 Lindeman, Mark. Rhode Island Presidential Risk-limiting Audit. 2020. https://elections.ri.gov/publications/Election_Publications/RLA/Rhode%20Island%20presidential%20RLA%20brief%20report.pdf.
States such as Rhode Island, operate their RLAs with a risk-limit of 9%, meaning that there is a 91% chance that an error would definitely be found and no more than a 9% chance that an error would not be caught by the RLA, although it could still be found by other checks not discussed in this paper.
8 Deluzio, Christopher. “A Smart and Effective Way to Safeguard Elections.” Brennan Center for Justice. Last modified July 25, 2018. Accessed January 10, 2021. https://www.brennancenter.org/our-work/analysis-opinion/smart-and-effective-way-safeguard-elections.
9 State of Colorado. Risk-Limiting Audit. By Post-Election Audit Initiative – Grant No. EAC110150E. https://www.eac.gov/sites/default/files/eac_assets/1/28/Risk-Limiting%20Audit%20Report%20-%20Final%20.CO.pdf.
10 “Post-election Audits.” National Conference of State Legislatures. Last modified October 25, 2019. Accessed January 10, 2021. https://www.ncsl.org/research/elections-and-campaigns/post-election-audits635926066.aspx.
11 Antonyan, Tigran, Theodore Bromley, Laurent Michel, Alexander Russell, Alexander Shvartsman, and Suzanne Stark. “Computer Assisted Post Election Audits.” State Certification Testing of Voting Systems National Conference, 2013. Accessed January 14, 2021. https://voter.engr.uconn.edu/voter/wp-content/uploads/AS-2013.pdf.
12 Risk-Limiting Audits Working Group. Risk-Limiting Post-election Audits: Why and How. https://www.stat.berkeley.edu/users/stark/Preprints/RLAwhitepaper12.pdf.
Ruth, David, and Amy Hodges. “Hand Counts of Votes May Cause Errors, Says New Rice U. Study.” Rice University News and Media Relations. Last modified February 2, 2012. Accessed January 10, 2021. https://news.rice.edu/2012/02/02/hand-counts-of-votes-may-cause-errors-says-new-rice-u-study/.
13 “Audit FAQ.” Verified Voting. Accessed January 9, 2021. https://verifiedvoting.org/audits/auditfaq/.
Risk-Limiting Audits Working Group. Risk-Limiting Post-election Audits: Why and How. https://www.stat.berkeley.edu/users/stark/Preprints/RLAwhitepaper12.pdf.
A 2012 UC Berkeley study found that the cost of manual RLAs is closer to around 51 cents each. However, RLA procedures have changed since this study was conducted and this figure may now be out of date. Thus, the authors chose to utilize the 2017 figure from Colorado.
14 “Current and past Election Results.” Georgia Secretary of State. Accessed January 29, 2021. https://sos.ga.gov/index.php/Elections/current_and_past_elections_results.
15 JBenaloh, Josh, Philip B. Stark, and Vanessa Teague. “VAULT: Verifiable Audits Using Limited Transparency.” E-Vote-ID, 2019, 69-90. Accessed January 13, 2021. https://www.e-vote-id.org/wp-content/uploads/2019/10/VAULT.pdf.
While there would likely be significant labor costs associated with software development, the majority of states would likely not have to pay it. From previous software developed in this realm, it is likely that the research and development would either be paid by the private sector or a single state. If it was paid by the private sector, the firm would then distribute it amongst states that purchase the software. If the software was paid for by a single state, it is likely upon completion the software would become open-source. This precedent was established by Colorado’s creation of RLAtool, a software used during the auditing process, which is now open-source.
16 Rodríguez-Pérez, Adrià, Pol Valletbó-Montfort, and Jordi Cucurull. “Bringing Transparency and Trust to Elections: Using Blockchain for the Transmission and Tabulation of Results.” Proceedings of the 12th International Conference on Theory and Practice of Electronic Governance (ICEGOV2019), 2019, 46-55. https://doi.org/10.1145/3326365.3326372.
17 Ibid. For a further breakdown and analysis of costs that are generally associated with blockchain but may not directly apply to this use case see above.
19 “The Verifier.” Verified Voting. Accessed January 29, 2021. https://verifiedvoting.org/verifier/#mode/navigate/map/ppEquip/mapType/normal/year/2020.
20 Antonyan, Tigran, Theodore Bromley, Laurent Michel, Alexander Russell, Alexander Shvartsman, and Suzanne Stark. “Computer Assisted Post Election Audits.” State Certification Testing of Voting Systems National Conference, 2013. Accessed January 14, 2021. https://voter.engr.uconn.edu/voter/wp-content/uploads/AS-2013.pdf.
Connecticut’s semi-automated audits allow for the incorporation of technology while still allowing for human interpretation of voter intent. This methodology has been found to reduce some of the barriers to implementing an audit, including labor intensity. While blockchain has a larger impact on reducing barriers to implementation, Connecticut’s semi-automated method is a feasible alternative for manually cast ballots.
21 Gerhart, Ann. “Election Results under Attack: Here Are the Facts.” Washington Post, January 14, 2021. https://www.washingtonpost.com/elections/interactive/2020/election-integrity/.
22 Nicas, Jack. “No, Dominion Voting Machines Did Not Delete Trump Votes.” The New York Times, November 11, 2020. https://www.nytimes.com/2020/11/11/technology/no-dominion-voting-machines-did-not-delete-trump-votes.html.
23Paxton Ken, Webster Brent, and Joseph Lawrence, “Motion to Leave to File Bill of Complaint” ( 2020). ;
Durkee, Alison. “17 States Agree the Supreme Court Should Overturn Biden’s Win.” Forbes. Last modified December 9, 2020. Accessed January 9, 2021. https://www.forbes.com/sites/alisondurkee/2020/12/09/17-states-agree-the-supreme-court-should-overturn-election-biden-win-texas/?sh=5dc3fb396452.
24 “Election Terminology.” United States Department of Commerce, National Institute of Standards and Technology. Accessed January 19, 2021. https://pages.nist.gov/ElectionGlossary/#voter-verifiable.
25 “Setting the Record Straight: Facts and Rumors.” Dominion Voting. Last modified January 29, 2021. Accessed January 29, 2021. https://www.dominionvoting.com/election2020-setting-the-record-straight/.
26 “Risk-limiting Audits.” National Conference of State Legislatures. Last modified February 17, 2020. Accessed January 10, 2021. https://www.ncsl.org/research/elections-and-campaigns/risk-limiting-audits.aspx.
27 Risk-Limiting Audits Working Group. Risk-Limiting Post-election Audits: Why and How. https://www.stat.berkeley.edu/users/stark/Preprints/RLAwhitepaper12.pdf.
“2020 General Election Risk-limiting Audit.” Georgia Secretary of State. Accessed January 10, 2021. https://sos.ga.gov/index.php/elections/2020_general_election_risk-limiting_audit.
Georgia’s November 2020 RLA successfully identified that there was a memory card issue which slightly impacted final vote counts. The correction of this error did not change the final results of the election, but still the identification of the error demonstrates the importance of RLAs.
28 Lindeman, Mark. Rhode Island Presidential Risk-limiting Audit. 2020. https://elections.ri.gov/publications/Election_Publications/RLA/Rhode%20Island%20presidential%20RLA%20brief%20report.pdf.
29 Burke, Peter J., and Jan E. Stets. “Trust and Commitment through Self-verification.” Social Psychology Quarterly 62, no. 4 (1999): 347-66. https://doi.org/10.2307/2695833.
30 Park, Sunoo, Michael Specter, Neha Narula, and Ronald L. Rivest. Going from Bad to Worse: From Internet Voting to Blockchain Voting. MIT, 2020. https://people.csail.mit.edu/rivest/pubs/PSNR20.pdf.
Specter, Michael A., James Koppel, and Daniel Weitzner. The Ballot Is Busted before the Blockchain: A Security Analysis of Voatz, the First Internet Voting Application Used in U.S. Federal Elections. MIT, n.d. https://internetpolicy.mit.edu/wp-content/uploads/2020/02/SecurityAnalysisOfVoatz_Public.pdf.
31 Utah County Elections Division. “Utah County Elections Live Voatz App Results Audit.” Facebook. Last modified September 4, 2019. Accessed January 10, 2021. https://www.facebook.com/1061986167328874/videos/375286443402581/.
32 Park, Sunoo, Michael Specter, Neha Narula, and Ronald L. Rivest. Going from Bad to Worse: From Internet Voting to Blockchain Voting. MIT, 2020. https://people.csail.mit.edu/rivest/pubs/PSNR20.pdf.
33 G, Mark, Melinda N, Lisa B, George Bell, Jonathan Downing, Kaivan Rahbari, Moh Kilani, Katlyn Woods, Scott S, Curtis T, Everett J, and Aaron Varrone. Blockchain and Suitability for Government Applications. 2018 Public-Private Analytic Exchange Program, n.d. https://www.dni.gov/files/PE/Documents/2018_AEP_Blockchain_and_Suitability_for_Government_Applications.pdf.
34 Nelson, Paul. Primer on Blockchain. United States Agency for International Development, n.d. https://www.usaid.gov/sites/default/files/documents/15396/USAID-Primer-Blockchain.pdf.
35 Bélanger, France, and Lemuria Carter. “Trust and Risk in E-government Adoption.” The Journal of Strategic Information Systems 17, no. 2 (2008): 165-76. Accessed January 13, 2021. https://doi.org/10.1016/j.jsis.2007.12.002.36 Lou, Antonio T. F., and Eldon Y. Li. “Integrating Innovation Diffusion Theory and the Technology Acceptance Model: The Adoption of Blockchain Technology from Business Managers’ Perspective.” International Conference on Electronic Business 4 (2017): 299-302. https://core.ac.uk/download/pdf/301374121.pdf.