Election day in New York’s June 22 primaries—the largest American city thus far to adopt ranked choice voting (RCV)—is a week away. While the method will be used in contests ranging from city council to borough president, the most high-profile race utilizing RCV is the mayoral race—and the stakes are high. After all, candidates are vying to lead a city of 8 million people, 300,000 government workers, and a budget of more than $95 billion.
But as New Yorkers continue to head to the polls, we know much less about the state of the mayoral race compared to previous years. There are probably two reasons for this: 1) recent scandals have caused some shifts not yet captured in all polls and 2) in comparison to previous mayoral races, as the Irish Times recently noted, “Polling of this race is limited, making it difficult to gauge the outcome.
Pollsters, citing the seemingly “unpredictable” nature of RCV polling, may have been wary of conducting such polls for fear of unfounded blows to their reputation, or simply because they are not sure how to best organize such a poll.
But other organizations have forged ahead, including Marist (which featured and cosponsored a recent poll with NBC NY), which recently released a comprehensive RCV poll that has Brooklyn borough president Eric Adams edging out former NYC sanitation commissioner Kathryn Garcia in the 13th round of voting, 56% to 44%. (FairVote also provided analytical support to Marist Polling with this poll). Their efforts—and those of other pollsters who have also released RCV polls—illustrate some best practices for conducting RCV polls.
Three of these practices essential to conducting any RCV poll (and which also make such polling much easier) are detailed below.
1. Reflect the ballot as much as possible
If this practice seems relatively simple, that’s because it is. In a race that uses ranked choice voting, pollsters should allow respondents to rank as many candidates as voters will be allowed to rank on their ballot—in New York’s case, five—and read the candidates in order of their appearance on the ballot. This method will simulate the voting process as best as possible and account for ballot appearance effects; i.e. a candidate receiving a higher share of the vote because they appear near the top of the ballot. Reflecting the ballot as it is is essential to conducting accurate RCV polling on any race.
2. Don’t require voters to rank every candidate
Unlike Australia’s form of preferential voting, no American adaptation of ranked choice voting requires compulsory ranking of all candidates. As such, to reflect the possibility that voters may only want to rank two or three candidates and not all five (in New York’s case), pollsters should give respondents the option to stop ranking after every new ranking. The recent NBC NY/Marist poll allowed voters to do just that, and a total of 74% of respondents chose to rank multiple candidates, and 24% ranked all five. This option accounts for the reality that, in practice, not all voters will rank all possible candidates, and, as such, produces the most accurate picture of an RCV race.
3. Acknowledge the limitations of confidence intervals
Without delving into the math, it is important to understand that, in RCV polling, confidence intervals—especially those reflecting lower first-choice support—will be relatively large, and this will affect results. For example, the NBC NY poll has Dianne Morales, Ray McGuire, and Shaun Donovan tied in last at 4% in the fifth round of tabulation, which is within the margin of error. But because RCV takes the candidate with the lowest support and redistributes the preferences of their voters to other candidates, the margin of error does not always reflect which candidate (in this case Shaun Donovan) actually has the least support among low-polling candidates. A best practice of RCV polling is to readily acknowledge this limitation so that the public can make accurate judgments from any poll.
Are you a pollster trying to navigate a ranked choice election? Reach out to us at [email protected]. FairVote can help navigate and explore survey results.