Privacy vs. Transparency: What a New Study Says About Releasing Ballots
Is an anonymous ballot no longer private once it is released to the public?
Across the country, policymakers are grappling with how to best report election results. Individual ballot records promise to increase the transparency of election results. But there are concerns about privacy.
According to a new assessment published today in Science Advances by a resident fellow at Yale’s Institution for Social and Policy Studies, the release of the more than 2 million ballots cast in the 2020 general election in Maricopa County, Arizona did not expose any of the vote choices made by 99.83% of the voters to possible public revelation. That’s compared with 99.95% of voters whose identity remains protected under the county’s current practice of reporting aggregate results by precinct and method of voting.
“Any election reporting system that promotes transparency has a risk of revealing how some people voted,” said ISPS faculty fellow Shiro Kuriwaki, who conducted the study with Jeffrey B. Lewis, professor of political science at the University of California, Los Angeles, and Michael Morse, Yale Law ’19 and assistant professor of law and political science at the University of Pennsylvania. “But in this case study we have shown that the risk of revelation is quite small, and the most aggressive releases don’t reveal much more than what many jurisdictions already do.”
Voting districts report the total votes for each candidate but do not necessarily release individual ballot records that could bolster public trust in election outcomes. For example, election officials in Alaska, Georgia, Maryland, and San Francisco release individual ballots. But the practice is inexistent or explicitly prohibited in states including Indiana, South Carolina, and North Carolina, where officials have cited threats to anonymity.
Kuriwaki, an assistant professor of political science who studies American elections, and his co-authors conducted the first empirical assessment to gauge the privacy risk of releasing individual ballots using an actual election.
Revelations can be publicly revealed when voters in the same precinct, voting on the same type of ballot (for example, voters on the same street in a precinct split into two districts), and using the same vote method (for example, an in-person ballot) all vote for the same candidate. In such a scenario, someone could determine how everyone voted by examining the public voter file of registered voters in that precinct assigned to the specific ballot style and recorded as voting in person. The authors did not conduct the revelation but showed the number of voters that could be revealed with such a method.
Such unanimous voting is more likely in a small precinct. And in a circumstance where one person, for example, voted for Joe Biden in a precinct where everyone else voted for Donald Trump, that one person could use publicly available information to deduce how the others voted. In this example, even an outside observer could predict with high probability how an individual voted.
The researchers found that if all individual ballots were released in Maricopa County’s 2020 election, at least one vote choice could be publicly revealed for less than a fifth of a percent of voters, amounting to about 3,500 voters, or about 2,400 more than could be revealed under the county’s current, less transparent practice.
Revelations are more likely for top-of-the-ticket offices than down-ballot races and depend on factors such as patterns of partisan segregation and people who hold unpopular views.
The researchers discuss the convenience and shortcomings of redacting some information in small precincts, which several jurisdictions do to allow for the release of ballots while reducing the likelihood of revealing individual vote choices. Kuriwaki said he cautions against intentionally introducing random noise into the results to protect privacy because such a practice, though useful in releasing other types of information for analysis, goes against the fundamental point of reporting election results, which is to determine with precision who won and by how much.
“I think the number of revelations we uncovered is small enough that it can be remedied through short-term or long-term solutions, such as redacting or redistricting,” Kuriwaki said, adding that officials should be sensitive about the people disproportionately vulnerable to revelation. “For example, someone who casts a provisional ballot in a small precinct where people vote more-or-less uniformly is repeatedly more vulnerable to revelation election after election.”
ISPS Director Alan Gerber, Sterling Professor of Political Science, who has collaborated on research examining voters’ attitudes towards the secret ballot, praised the paper for its measured examination of a salient topic.
“Concerns about voter fraud and election security are only growing in the United States and abroad,” Gerber said. “Rigorous and timely academic studies like this one can help address these issues and restore public confidence in democracy.”