Tabulating Partial Ranked Choice Voting Results
Will New Yorkers have to wait for weeks for the results of the June primaries? Making ranked choice ballot data available as the votes are counted will allow for earlier result projections and provide clarity in a majority of the primary races.
***Update 5/15: I made a video to explain in detail how votes are counted in a Ranked Choice Voting election.
In June, NYC will switch to Ranked Choice Voting for its primary elections for public advocate, comptroller, borough president, city council, and mayor. Over 50 primary races will be decided based on the ranked preferences of voters. Ranked Choice Voting (RCV) has gained traction nationwide because of several advantages: it is seen as a fairer way to run elections, it avoids the “spoiler effect,” and saves money by avoiding runoff elections. However, it also has some drawbacks: voters may be confused by the ranking mechanism — several voter education campaigns are underway — and the election results take a long time to be processed as the vote transfers are only tabulated once all votes are in.
RCV got a trial run in NYC earlier this year with three special elections, which like primaries moved to RCV in 2021. (A special election for Queens City Council 24th would have been the first one to use RCV but one candidate received over 50% of the votes in the first round so no RCV rounds were necessary.) The first one, for Queens City Council 31st, was held on February 23rd, but the official results, with the full RCV tabulation, were only certified on March 19th. The second and third special elections, Bronx City Council 11th and 15th, were held on March 23rd, and results were available on April 15th. For over three weeks, the only information given to the public for all three elections were the election night results (see images below), which gave the ranked choice first-place numbers for each candidate, but did not provide any information as to how the ranked votes transfers may look like.
These lengthy delays in elections with a relatively small number of votes suggest that the results of the June primaries, which include several citywide competitive races and are likely to receive many more absentee and mail-in ballots, will not be known until the second half of July or later, unless the NYC Board of Election (BOE) adds more transparency to the process so that news media, researchers, and other independent entities can make result projections.
Election results being certified weeks after the election is common, as absentee and mail-in ballots need to be counted, challenged votes cured and all data verified. In most traditional simple majority (or plurality) elections however, the winner can be identified with some votes still outstanding as information on the tallied votes can provide an estimate of the results. Vote counting in RCV election is more complex as the order of vote transfers (i.e., the order in which candidates are eliminated) may change the final outcome. For the three NYC special elections, the actual RCV rounds counting was done manually, with paper ballots being moved from pile to pile. This process took several days for the Bronx City Council 11th and 15th elections, which had relatively low turn out. The NYC BOE is planning to use a software for the June primaries, which would speed up the tabulating process from several days to minutes, but all votes would still need to be fully accounted for before the electronic RCV transfer tabulations can start. The results will still take weeks to be produced.
If the NYC BOE were to make known partial cast ballot data available as votes are processed, valuable information could be inferred from the ballots, as early as election night, that may help make earlier projections of whom the eventual winner will be.
What can we learn from partial cast ballot data?
The challenge with RCV counting is that as candidates are eliminated, and votes are transferred, the order of candidates may change. The candidate with the most 1st choice ranked vote in round one may not be the top candidate in round 2, or 5, or in the final round. In practice, this does not happen that often; a study by Fairvote of all RCV races in the US since 2004 shows that in the vast majority of cases, the first round leader is the overall winner of the election. In about half of the races, the first round leader received more than 50% of the votes, making the RCV vote transfers unnecessary. Out of the remaining 124 elections that required several rounds of RCV counting, 15, or 12%, had a “come-from-behind” winner.
The question is how soon can we identify who is the most likely winner of the race? Can we use statistical and deductive methods to predict the winner before all the votes are tallied?
- Deductive reasoning: Partial voting results can sometimes provide enough information to eliminate candidates before all votes are known. The vote transfers from these candidates can then be counted, providing a better idea of who the final contenders are.
Looking at the Bronx City Council 11th election night results (left image above), with 6,994 votes and 97.59% of scanners reporting , we have 173 still missing from scanners. Assuming these include all the votes cast in the election, we can see that even if lowest first votes candidates Carlton Berkley or Kevin Pazmino were to win all first choice votes on these missing ballots and receive respectively 308 or 345 first choice votes, neither could overcome the next lowest vote candidate Daniel Padernacht, with 899 known first-choice votes. One of them will be eliminated first, but we cannot differentiate them without tabulating more votes as the vote difference between them is less than that of the unknown votes. We can however observe that regardless of the order of their elimination, they will both be eliminated in the next two rounds: even if their votes fully transfer from one to the other, and all unknown votes and all write-ins votes transfer to the same candidate, the maximum number of votes that candidate could get in the next RCV round is 503 (135+172+23+173), not enough to jump ahead of Daniel Padernacht. We can therefore conclude that both will be eliminated next and look at their votes transfers right away, without waiting for the yet-to-be-counted votes. The information gathered from the vote transfers can then be used for further deductive reasonings, and potentially eliminate more candidates and transfer their votes. - Branching time logic: Simple deduction can help reduce the field of candidates, but in most cases it will not be enough to gain sufficient insights on the results of the election. In the example above, the actual election had a significant higher number of uncounted votes, likely from absentee ballots, so the deductive process would not have allowed for meaningful winnowing of the candidate field from election night results.
Using temporal logic, we can draw inferences and make predictions by looking at the available data and simulate what-ifs outcomes. What if Carlton Berkley is eliminated first? What if Kevin Pazmino is eliminated first? By investigating all possible futures based on known data (the full ranked lists of known ballots), we can identify which election outcomes are possible, and which are not. The expected distribution of yet-to-be-counted votes can be included in a probabilistic analysis of possible futures to make results predictions. - Condorcet winners: Partial cast ballot data would also allow us to study all pairs of candidates and see which candidate would be the winner of all possible runoff elections (the last round of RCV is de facto a runoff election). The pairwise matchups results can easily be extracted from the RCV ballots: if candidate A is ranked higher than candidate B (or if B is not ranked) A wins one vote in the matchup, if neither is ranked we treat the matchup vote as an abstention. An advantage of a pairwise matchup based analysis is that there are many well understood techniques to estimate the result of a two candidate runoff election from partial results that can be applied to estimate the result of each pairwise election.
In Voting theory, a Condorcet winner is a candidate that wins all pairwise matchups agains all other candidates. Not all RCV elections will have a Condorcet winner, and Condorcet winners are not guaranteed to win their RCV elections: because the elimination rounds only look at first-choice placements, there may be a candidate who wins all their one-on-one matchups but is eliminated before the final round as they don’t have enough first choice votes (e.g., a candidate who is everybody’s second choice). The Fairvote study shows that it is extremely rare: out of the 322 RCV elections in the US since 2004, only one Condorcet winner did not end up being the final winner. Calculating partial pairwise election results can give us insights on the potential election outcome; identifying the Condorcet winner of an election can help identify come-from-behind winners.
San Francisco RCV data
San Francisco, which switched to RCV in 2004 for many of its local races, has been providing the public with daily result updates, including cast ballot images, provided as encoded text data, or more recently in Cast Vote Record (CRV) format. Once NYC moves to an electronic system for tabulating RCV results, there is no reason for the BOE not to make the cast ballot data public starting on election night, and with regular updates until the results are certified.
How can NYC improve the process in time for the upcoming election?
The NY City Council is currently considering a Local Law proposal that would require the New York City Board of Elections to publish the unofficial election night results for RCV elections. The Law purports to be designed to “add transparency to the RCV tabulation process and give voters confidence in the legitimacy of RCV election results”. Unfortunately, as of now, the text of the proposal only asks for first-votes totals or RCV round-by-round tabulations. As discussed above, this would not provide enough fine-grained information to fully understand the unofficial tallies: either option would hide a lot of information in the aggregates, resulting in information loss, less transparency, and less potential for analyzing and understanding the results of the election and the voting patterns.
To ensure a transparent voting process, and to allow for a better understanding of voting patterns and early results projections, the City Council, and New Yorkers, should demand that the NYC BOE provide cast ballot data for all tallied votes as soon as election night.
