Reporting Ranked Choice Voting Election Results with Outstanding Ballots
In 2021, I wrote a post highlighting potential issues with delayed reporting of Ranked Choice Voting (RVC) elections in NYC due to the late counting of absentee ballots. Since then, I have worked with Alborz Jelvani, a Rutgers CS graduate student, on algorithms for identifying possible winners in RCV elections when some ballots are still outstanding. Our work was published at the 10th AAAI Conference on Human Computation and Crowdsourcing (HCOMP 2022); I give an overview of our findings in this post.
Ranked Choice Voting (RCV) is used in national elections in several countries, such as Australia, Ireland, and the United Kingdom.
In the U.S. multiple counties and municipalities have recently adopted RCV for their elections with positive impacts on voter participation. One such municipality is New York City, which holds primaries using RCV for a large number of local elections; its most high-profile being the Mayoral Democratic Primary, which in June 2021 was all but guaranteed to determine the eventual NYC Mayor in a city that leans heavily Democrat.
Ranked Choice Voting has gained traction because of several advantages: it is seen as a fairer way to run elections and to improve the representation of women, voters of color, and participation in youth voters. It avoids the “spoiler effect,” reduces “wasted” votes when many candidates are running and saves money by avoiding runoff elections. In addition, by aiming at forming a consensus behind the selected candidates, RCV decreases incentives for strategic voting.
A main drawback of RCV is that the round-by-round process requires all the ballots to be tallied before the results of an election can be calculated. With increasingly large portions of ballots coming from absentee voters, RCV election outcomes are not always apparent on election night and can take several days or weeks to be published, leading to a loss of trust in the electoral process from the public. In 2018 the San Francisco mayoral results took a week to be tabulated and confirmed in large part due to the late counting of mail-in ballots. In NYC, the June 2021 primary results were certified a full month after the election due to the large number of absentee ballots; preliminary RCV results were not made available to the public for over a week after the election and did not originally include absentee ballots. These delays, the lack of transparency, and the incomplete information, or lack thereof, on the outcome of cast ballots, lead to distrust in the RCV election process from a population that is used to having election results, or at least close estimates, soon after an election.
To improve transparency in the reporting of RCV election results, Alborz Jelvani, a Rutgers CS graduate student, and I presented an algorithm to process partial results of RCV races without requiring all votes to be gathered before the counting can start at the 10th AAAI Conference on Human Computation and Crowdsourcing (HCOMP 2022). Our approach considers known voter preferences from tallied ballots to identify possible elimination orderings and takes into account the uncertainty associated with still-missing (e.g., absentee) ballots. Our algorithm allows for identifying candidates who still have a path to victory, and those who do not, as soon as election night, providing stakeholders with more transparency on the election outcome. The details of the algorithm can be found in our paper; in the rest of this post, I will highlight some of our results on data from the June 2021 Primary election.
Election Night Results of the June 2021 NYC Primary
Our results show that it would have been possible to call winners for 21 of the 52 NYC primary contests that had more than two candidates as soon as election night. Half of these would have been obvious even without our algorithm as the winner had over 50% of first-choice votes, but for the other half, the winner wasn’t necessarily clear until weeks after the election when the full RCV counts were produced. In some cases, our algorithm was able to identify results even when the election night results were very close. For instance, in the NYC 40th City Council District Primary, we were able to identify the only possible winner even though she only had 25% of first-choice votes on election night and two other candidates had 20% each.
The next four races illustrate how our algorithm can help provide better information as early as election night:
June 2021 NYC Mayoral Democratic Primary
The results of the Mayoral Primary were understandably the most awaited results of the primary. That particular race turned out to be a perfect illustration of the benefits and drawbacks of RCV. On election night, Eric Adams was leading with 31.66% of ballots, Maya Wiley was second with 22.22%, and Kathryn Garcia third with 19.58%. Andrew Yang was a distant fourth with 11.66%. The data made public on election night was limited to first-choice votes, so the outcome of the race was uncertain. It was clear that vote transfers would decide the election result and could lead to surprises. A week after the election, a count of RCV was performed only on in-person votes, omitting absentee ballots. The outcome showed that once Andrew Yang was eliminated in the third-to-last round, his vote transfers were enough for Kathryn Garcia to edge out Maya Wiley by only 400 votes, only to lose to Eric Adams in the last round.
These partial results raised more questions than they answered: over 165,000 absentee ballots were still to be counted. What if Maya Wiley were to be in the final round against Eric Adams? Would she win against him? How would votes transfer in other possible election orders? Could the final two be Maya Wiley and Kathryn Garcia? What would happen in that scenario?
Possible Elimination Orders for the 2021 NYC Democratic Mayoral Primary
Our algorithm would have been able to answer some of these questions: the figure above shows the possible elimination orders for the mayoral primary. The elimination paths can be read top to bottom, one round (level) at a time, with the initials corresponding to that of the eliminated candidate in each round. All three candidates, Eric Adams, Kathryn Garcia, and Maya Wiley had paths to victory (bottom row), but Eric Adams was guaranteed to finish first or second.
Our algorithm also provides more insights: it identifies that Maya Wiley needed a minimum of 66,440 (40%) absentee ballots votes for a path to victory while Kathryn Garcia would have needed a minimum of 15,776 (9.5%). Eric Adams had a path to victory that did not require him to win any absentee ballots (such as an unlikely scenario where all absentee ballots are shared among minor candidates). This would have shown that all three were potential winners, but Maya Wiley had to appear before the other two candidates in a much larger number of absentee ballots to have a chance at winning.
June 2021 Kings Democratic Member of the City Council 36th Council District Primary
On election night, candidate Tahirah A. Moore was tied in second place with Henry L. Butler (each at 4720 and 4721 ballots respectively) in the first round. The first-place candidate (and eventual winner) Chi A. Osse was ahead of them by 2969 ballots. Our algorithm identifies Tahirah A. Moore as the only other possible winner, despite Henry L. Butler having the same number of votes when 1883 absentee ballots were present (approximately 8% of all the cast ballots). Moore needed a minimum of 1726 absentee ballots for a possible path to victory. Therefore, on election night our algorithm could practically identify the winner in this contest as it is unlikely Moore would appear above Osse in 92% of the absentee ballots. Interestingly, the partial RCV count reported by NYC’s Board of Election a week after the election, using only in-person votes, shows Osse and Butler as the final two candidates, as it explores only one possible elimination path.
June 2021 Kings Democratic Member of the City Council 35th Council District Primary
On election night, candidate Michael Hollingsworth was in second place, closely trailing leading candidate (and eventual winner) Crystal Hudson by 1291 ballots in the first round. Our algorithm identifies Hollingsworth as the only other possible winner, needing a minimum of 2121 absentee ballots for a possible path to victory. However, there exist only 3089 absentee ballots; it seems unlikely that Hollingsworth would receive over 68% of the missing ballots and have a path to victory. Therefore, by election night, we could have inferred that there was likely only 1 winner for this contest even though Hollingsworth and Hudson had 34.45% and 38.49% of the election night ballots respectively.
June 2021 New York Democratic Member of the City Council 9th Council District Primary
This is one of three races from the NYC 2021 Primary with a come-from-behind winner, i.e., a winner who did not have the largest number of first-choice votes. While Bill Perkins had 21.1% of first-choice votes, ahead of Kristin Richardson Jordan’s 19%, once all ballots were tallied and the RCV rounds processed, Richardson won the race. Our algorithm identifies both Richardson and Perkins as possible winners and further identifies that both have paths to victory without needing to pick up absentee ballots depending on the order in which other candidates are eliminated.
Our experiments were performed on NYC cast ballot data that was only made public months after the June 2021 Primary election. Our work shows that it is possible to increase transparency and trust in RCV election processes by using reporting tools that focus on increasing clarity to voters and candidates on the election outcome if election night data is made public immediately. This should prompt localities that use RCV as their election mechanism to identify all possible winners when reporting preliminary results and to make cast ballot data available to the public as soon as possible for transparency.
