The Ticket Splitting Visualizer

Principal Investigator: Shiro Kuriwaki

(Assistant Professor of Political Science, Fellow at the Institution for Social and Policy Studies at Yale University)

About

The Ticket Splitting Visualizer transforms crowd-sourced ballot records into an interactive platform showing how voters make choices between various congressional, state, and local contests as they work through a ballot. This interface is the first of its kind to use actual ballots to show, for example, exactly how many Trump voters voted to approve sales taxes in their own city, or how many Biden voters voted for the Republican Marjorie Taylor Greene on the same ballot. Understanding for which candidates voters split their ticket in this way and by how much is an enduring question for election observers, including the politicians themselves.

These insights come at a critical time for how elections are run in America. Many voters are unsure whether their votes are accurately counted, and after the 2020 election, that ambivalence has sometimes turned into the hostile questioning of the integrity of elections. When a candidate significantly outperforms or underperforms a same-party candidate running in a different office, for example, some suspect that ballots were manipulated to engineer specific winners. In reality, though, voters can and do split their ticket for candidates of different parties quite often. The project is a testament to how transparent election technology in the United States can be, despite its real need for financial and political support.

FAQ

What am I looking at in these tables?
Where does the ballot data come from?
How do we know the data is legitimate?
Are these representative of the Georgia election?
Why Georgia?
Is this data public record?
Does the visualizer violate voters' privacy?
What about 2024?
Can I download the ballot-level data to do more complex analyses?
How do I cite this data?

What am I looking at in these tables?

The Ticket Splitting visualizer lets you explore how voters actually voted on two contests on the same ballot. You pick two contests to compare: one contest to put on rows, another to put on columns. The visualizer will then show voting patterns among voters who were given those contests on the ballot.

A Ticket Splitting Visualizer pivot table showing votes for U.S. President compared with votes from a Local Sales Tax Referendum

Users should then display the number of voters as percentages. The example below compares the votes for US President with the Gwinnett County Sales Tax Referendum, which extended the special local sales tax for education purposes by 1% until 2027. By showing these as ROW percentages, each row sums to 100%. 83.8% of the 159,906 Biden voters in this data voted Yes, while 13.6% voted no, and 2.6% undervoted (that is, turned out to vote but skipped this contest). Selecting COLUMN percentages will show, in contrast, the percentage of Yes supporters who voted for Joe Biden. All these formats show the same number of voters, but expressed as different fractions.

How is this different from traditional election reporting?

Usually, election results are reported in tallies for one contest at a time. In other words, they reveal the column totals and row totals of the ticket splitting visualizer, but not the internal, within-voter patterns. It turns out that even if a candidate in one contest receives 50% of the vote and a candidate in another contest also receives 50% of the vote, the number of ticket splitters could range from 0% to 100% of the electorate. The Ticket Splitting Visualizers shows exactly how much.

Where does the ballot data come from?

We originally obtained records from a public website maintained by Jeffrey O'Donnell, who leads what Bloomberg News called "a persistent group of election skeptics." This group makes public records requests in their local areas for cat vote records. O'Donnell refers to himself and his collaborators as the Raccoon Army. My academic collaborators and I independently inspected the original public records files they have posted, checking if we could reproduce certified election tallies.

Dozens of researchers worked with me to analyze and format the data. In particular, my collaborator James M. Snyder (Harvard) did a significant share the manual coding of local candidates and ballot measures by searching newspaper coverage. I have also collaborated extensively with Jeffrey B. Lewis (UCLA) and Mason Reece (MIT) to process and analyze this data. My other collaborators in this project are listed in the references below. I also thank the election administrators. I did not send public records requests to local election offices for this 2020 project, in part to respect their main duties to maintain voter rolls and prepare elections for their constituencies. But all the data here originate from the local election official's response to public records requests.

How do we know the data is legitimate?

The data presented here is not an audit of the election. Cast vote records may not include all of the ballots cast in an election for several reasons: sometimes valid ballots are held aside for manual adjudication and are not scanned through tabulators, and therefore a cast vote record is not created (but the vote is counted in the certification later). Also, some of the cast vote records released to the public and posted on their website are copies of the actual cast vote record, which are modified with minor privacy-protecting redactions.

Because O'Donnell's website is anonymous and we researchers are not part of the chain of custody, the data may have been corrupted before posting without our knowledge. That said, my collaborators and I strengthened our confidence that CVR files posted are uncorrupted and genuine by comparing vote tallies produced using the downloaded ballots data to the official reports of vote totals from the same jurisdictions. Read our peer-reviewed article in Nature Scientific Data for our methodology and results, as well as for an explainer for cast vote records: Cast vote records: A database of ballots from the 2020 election.

Are these representative of the Georgia election?

No. 2020 Turnout in Georgia was about 5 million votes, while we only have CVR data from 4.2 million votes, covering 89 of the 159 total counties. In this subset, Biden won 51.6 percent of the vote while he won 50.1 percent statewide. In other words, the subset of counties we study leans one-percentage point more Democratic than the whole state.

Why Georgia?

Georgia is one of 24 or so states in the Raccoon Army collection of cast vote records and our political science journal articles. We chose to focus on Georgia in this ticket split visualizer because of its wide coverage, two U.S. Senate races, and multiple local ballot measures and school board races. We are looking to expand the visualizer to more states and years.

Is this data public record?

In short, yes. We presume all data posted by the Raccoon Army was obtained through valid open records requests, as they report on their website.

Does the visualizer violate voters' privacy?

No. In this ticket splitting visualizer, we do not allow users to download individual ballots with precinct identifiers. In an academic article with Jeff Lewis and Michael Morse, I show that revelation risk of cast vote records can be comparable to that of more standard precinct-level returns. Read more about the legal and policy implications of the secret ballot in our article, "Privacy Violations in Election Results".

What about 2024?

Various individuals, including O'Donnell and my collaborators, are collecting 2024 cast vote records.

Can I download the ballot-level data to do more complex analyses?

The Georgia data that we cleaned, reformatted and evaluated (and which underlies this website) is deposited at Dataverse at DOI: 10.60600/YU/N4SXIO. See our Nature Scientific Data article for how the dataset is structured and how to read it in via the arrow software. O'Donnell's website is https://votedatabase.com.

How do I cite this data?

Please cite our article, How Partisan are U.S. Local Elections? Evidence from 2020 Cast Vote Records," forthcoming in the American Political Science Review. This article analyzes all state and local contests.

Contact: If you have any comments or suggestions, please feel free to reach me at shiro.kuriwaki@yale.edu.

Disclaimer: The statements and errors in this website are mine only and do not reflect those of election administrators, my coauthors, or my employer. Yale University does not generally endorse research findings. Our use of the public records originally requested by the Raccoon Army does not reflect any endorsement of the Raccoon Army's interpretation of the data, either.

Cited Research

Atkenson, Lonna, Eli McKown-Dawson, M.V. Hood III, and Robert Stein (2023). "Voter Perceptions of Secrecy in the 2020 Election". Election Law Journal
Bloomberg Technology (2022). "Raccoon Army Swamps Election Officials in Dubious Campaign to Disprove Results." October 25, 2025. https://perma.cc/J2H6-TAUD
Conevska, Aleksandra, Shigeo Hirano, Shiro Kuriwaki, Jeffery B. Lewis, Can Mutlu, and James M. Snyder, Jr. (2025) "How Partisan are U.S. Local Elections? Evidence from 2020 Cast Vote Records." Accepted Conditional on Replication, American Political Science Review.
Gerber, Alan S., Gregory A. Huber, David Doherty, and Conor M. Dowling (2013). "Is there a secret ballot? Ballot secrecy perceptions and their implications for voting behaviour." British Journal of Political Science 43 (1).
Holliday, Derek E., Justin Grimmer, Yphtach Lelkes, and Sean J. Westwood (2025)."Who Are the Election Skeptics? Evidence from the 2022 Midterm Elections." Election Law Journal
Kuriwaki, Shiro (2025). "Ticket Splitting in a Nationalized Era." Journal of Politics.
Kuriwaki, Shiro, Mason Reece, Samuel Baltz, Aleksandra Conevska, Joseph R. Loffredo, Can Mutlu, Taran Samarth, Kevin E. Acevedo Jetter, Zachary Djanogly Garai, Kate Murray, Shigeo Hirano, Jeffrey B. Lewis, James M. Snyder Jr., and Charles H. Stewart III (2024). "Cast vote records: A database of ballots from the 2020 U.S. Election." Nature Scientific Data 11, 1304.
Kuriwaki, Shiro, Jeffrey B. Lewis, and Michael Morse (2025). "Privacy Violations in Election Results." Science Advances 11 (11)
New York Times, "A Republican Election Clerk vs. Trump Die-Hards in a World of Lies" (2024). June 6, 2024.
Reece, Mason, Joseph R. Loffredo, Alejandro Flores, Samuel Baltz, and Charles H. Stewart III. (2024+). "Hidden Partisanship in American Elections." SSRN.
Stewart, Charles H. (2022). "Trust in Elections." 151 (4) Daedalus

Don Blankenship

Opponents

Received at least 1% of vote