RCV simulation model

Simulation Model Overview

In addition to the basic ranked choice voting tabulation above, we ran an RCV simulation model for city council district 50 to explore other possible outcomes and account for uncertainty in the race, especially given the uncertainty about the second-place finisher.

Our model confirms David Carr’s position as a likely winner with a narrow margin. Carr leads in 63% of our simulations, with Pirozzolo winning 33% of the time, and a Kepi victory 3% of the time. 

Most likely final-round pairing: Carr vs. Pirozzolo

The most likely final-round match-up in our simulations is Carr vs. Pirozzolo, a matchup which occurs 80% of the time. Carr wins most of those match-ups with a median margin of 52%-48%. 

A Carr vs. Kepi final round is also possible, occurring in 13% of our simulations. In those scenarios, Carr typically leads by a similar margin.

RCV simulation methodology

Our RCV Simulation is based on 10,000 bootstrapped RCV tabulations created from weighted poll responses. Each simulated RCV tabulation samples from the distribution of weighted poll responses and tabulates the ranked choice voting tally based on that sample. 

Head-to-head comparisons

We use ranked preferences to simulate head-to-head match-ups to offer insight into which candidates would fare best against other candidates head-to-head. A candidate ranked higher than another candidate by a survey respondent is assumed to win that head-to-head match-up.

All ranked candidates are assumed to be preferred to non-ranked candidates.

Percentages below show the portion of head-to-head match-ups won by the candidate in the first column compared to the candidate in the top row. Read the first line with data as, “Carr wins 57% of match-ups against Kepi, 52% vs Pirozzolo,” and so on. 

Carr leads head-to-head match-ups against every other candidate. Pirozzolo leads against all but Carr. 

Impact of RCV

• Respondents rank more than 3 candidates on average. Respondents demonstrated a strong use of rankings, ranking an average of 3.3 candidates out of 4 in the borough president race and 3.2 candidates out of 5 for the city council race. Middle-aged respondents (35-65) ranked the fewest while younger and older respondents ranked the most candidates. Those aged 65 and older used the most rankings.  
• Ranked choice voting makes some respondents more likely to vote. When asked whether the opportunity to rank candidates makes them more or less likely to vote in theJune 22 primary, a majority of respondents (60%) said that it had no impact. However, 21% said it made them more likely to vote, a 50% increase over the 14% who said less likely. The biggest impact is among Hispanic respondents, where 36% said RCV would make them more likely to vote compared to 8% who said less likely.  
• Respondents say ranking is easy. When asked whether they would describe the process of ranking as easy or hard, 70% of respondents found ranking to be easy, 16% found it hard or very hard, and 13% were not sure.  
• Most respondents are familiar with RCV, but some gaps exist. 61% of respondents say they are familiar or very familiar with RCV, while 35% are not familiar and 3% are not sure. This is a lower level of familiarity than among Democratic respondents, 77% of whom are familiar with RCV based on previous Citizen Data polling. There were many more Democratic primaries using RCV in 2021 than Republican primaries, possibly explaining the discrepancy. Among these Republican respondents, the group with the highest familiarity is Hispanic respondents at 69%. 

Staten Island borough president race

This poll also asked respondents for their ranked preferences for Staten Island Borough President. In this case, our poll results differ significantly from the first-choice preferences reported on election night.

RCV TALLY SCENARIO 1: RAW POLL DATA ONLY

This poll underestimated support for Vito Fossella by 14 points, while over-estimating support for every other candidate by 2-5 points. Given the accuracy of Citizen Data polling in the Democratic mayoral primary and the district 50 city council race above, we hypothesize that Fossella’s improved performance is due to Donald Trump’s late endorsement, moving voters towards Fossella in the final days of this campaign which this poll did not capture. To explore the polling data given the unexpectedly strong first-round performance for Fossella, we re-run the RCV tabulation based on actual first choices rather than first-choice preferences from the poll. In this scenario, if we hold everything else constant, Matteo would have a narrow advantage because he performs better as a back-up choice among Remauro supporters. 

RCV TALLY 2: INCORPORATE ACTUAL FIRST CHOICES, ALL ELSE CONSTANT

However, we might reasonably assume the poll under-counted Remauro -> Fossella transfers to a similar degree as it under-counted Fossella first-choice preferences. As a third scenario, we run the RCV tabulation assuming that when Remauro is eliminated, Fossella earns a 14-point advantage over Matteo compared to the polling average. Using these hypothetical transfer probabilities, a third RCV tally reveals a narrow lead for Fossella in the final round. 

RCV TALLY 3: INCORPORATE ACTUAL FIRST CHOICES AND ADJUST TRANSFER PROBABILITIES