Polling Ranked Choices for the 2026 Maine Primaries

Executive Summary

Since 2018, Maine has used instant-runoff elections with ranked-choice ballots (often called simply “ranked-choice voting opens in a new tab”) for all statewide primaries. With the 2026 Maine primaries opens in a new tab quickly approaching on June 9, the latest University of New Hampshire poll opens in a new tab shows that two primary races, the Democratic Gubernatorial and Democratic ME-02 House-seat primaries, are both in apparent dead heats. For the governor’s seat, the latest poll shows Troy Jackson gaining considerably on longtime frontrunner Nirav Shah, with Jackson now receiving 28.4% of first-choice votes, compared to Shah’s 28.0%. Meanwhile, in the ME-02 House seat Democratic primary, Jordan Wood barely leads a crowded field to face Republican candidate Paul LePage, with 24.5% of first-choice votes for Wood compared to 22.0% for Baldacci, 16.7% for Matthew Dunlap, and 13.6% for Paige Loud.

With such close polling margins in first-choice votes and no candidate reaching a clear majority, the mechanics of ranked-choice voting make it challenging to forecast the outcome: as candidates get eliminated sequentially in the instant-runoff process, the transfer of votes from them to other candidates becomes critical in deciding the winners of these two primary elections. While many polls for ranked-choice elections only ask about first-choice votes, these latest UNH polls are, fortunately, ranked-choice. As researchers studying ranked-choice voting, these polls thus provide unique opportunities to study such polling for competitive elections and to investigate how reliably such polls can forecast the outcome of ranked-choice elections. Analyzing these polls, we find:

1. In the Democratic gubernatorial primary, despite the recent cross-endorsement amongst Troy Jackson, Shenna Bellows, and Hannah Pingree, we find that Nirav Shah still receives many vote transfers, even from voters who rank Bellows or Pingree first. As a result, even though the first-choice poll results are very close, when simulating instant run-off elections based on the polls, Shah wins nearly all simulated primaries (97.4%).
2. However, given Jackson’s strong support among younger voters, whose turnout has been rising, and the recency of the cross-endorsement, it remains possible that preferences will shift ahead of the June 9 primary. In simulated elections that yield a Jackson victory, Jackson has strong first-choice shares to offset vote transfers.
3. In the ME-02 House Democratic primary, meanwhile, the dead heat between Wood and Baldacci in first-choice votes persists in election simulations. In our analysis, Baldacci garners greater support from vote transfers than Wood, winning 67% of simulated primaries while Jordan Wood wins 20.5%, and Matthew Dunlap wins 12.5%. Paige Loud does not win any of our simulated primaries. Most notably, the candidate who wins the highest percentage of our simulated elections, Baldacci, is not the candidate with the highest first-choice vote count (which is Wood).

Overall, our analysis of the UNH ranked polls in these two ranked-choice primaries provides striking examples of the complexity of forecasting competitive ranked-choice elections. While one seemingly close election (the gubernatorial race, based on first-choice polls) is actually not so close, the other seemingly close election (the ME-02 House race) is actually incredibly close, with three candidates all having sizable probabilities of winning the election in our simulations.

Our analysis is based on the top-3 ranked choices collected from the UNH Survey Center poll, the largest poll of Maine voters ahead of the primaries, and one of only two non-partisan polls opens in a new tab. It was collected May 21-25, and the UNH analysis opens in a new tab was released May 27th. In our analysis, we drill into these two Democratic primaries because the Republican gubernatorial and ME-02 primaries are far less competitive, with Paul LePage running uncontested in the Republican ME-02 primary and Robert Charles in a commanding lead in the GOP gubernatorial primary. Further details about our simulation can be found in the Methodology section, and questions and comments can be directed to David Liu at davidliu@cornell.edu opens in a new tab.

Democratic Gubernatorial Primary

With incumbent Janet Mills reaching the gubernatorial term limit, a crowded list of candidates declared for the 2026 primaries opens in a new tab, five for the Democratic primary and eight for the Republican one. For much of the race, Nirav Shah, who led Maine’s CDC through the COVID-19 pandemic, has steadily held the lead in first-choice polls opens in a new tab. However, Jackson’s surge in the latest poll (dead even at 28% with Shah for first-place votes), accompanied by endorsements from Senator Bernie Sanders opens in a new tab and major labor unions opens in a new tab, demands a closer analysis of vote transfers beyond first-choice preferences.

The candidates themselves have recently begun expressing how they hope their supporters will rank. For instance, Jackson, Hannah Pingree (the current speaker of the Maine House of Representatives), and Shenna Bellows (the current Secretary of State of Maine) cross-endorsed opens in a new tab each other late in the race on May 22, telling their supporters to rank the others second and third opens in a new tab. The strategic move takes ranked-choice voting into consideration and aims to consolidate support for the self-identified “Progressive” candidates through vote transfers. The announcement parallels the alliance between Kathryn Garcia and Andrew Yang opens in a new tab in the 2021 New York City Democratic primary, when the two aligned against front-runner Eric Adams, as well as the cross-endorsements between Jane Kim and Mark Leno opens in a new tab in the 2018 San Francisco mayoral special election, when both sought to overtake front-runner London Breed. With the May 22 Jackson-Pingree-Bellows cross-endorsement announced midway through the UNH poll's data collection period (May 21-25), it remains unclear whether voters will coalesce further around the three candidates (increasing vote transfers among them) ahead of the June 9 primary.

Bellows and Pingree Voters Transfer to Shah

In the UNH poll, we find that Bellows' and Pingree’s supporters frequently rank Shah over Jackson. If the recent cross-endorsements change these down-ballot ranks, it could thus have a considerable effect on the outcome of the election. The chart below shows the probability that each candidate is ranked second, given a first-choice candidate. Those who rank Bellows first rank Shah second 32.6% of the time, compared with 24.1% (-8.5%) for Jackson. Similarly, those who rank Pingree first rank Shah second 29.3% of the time compared to only 9.9% (-19.4%) for Jackson.

Forecasting the Democratic Gubernatorial Primary

We use the top-3 UNH poll, combined with a model of ranked choices, to simulate how the election might play out (see the Methodology section for details). In simulated elections, the transfer of votes to Shah leads to consistent victories for the longtime front-runner despite the closeness in first-choice selections. The chart below summarizes the results of all simulated Democratic Gubernatorial elections. Each point on the plot is a simulated election. The x-axis is Shah’s margin in first-choice votes, whereas the y-axis is Shah’s margin in vote transfers from all other candidates. The dashed line is the victory boundary, denoting the boundary along which the first-choice and transfer margins cancel each other out.

Although the simulated elections are nearly evenly divided on first-choice votes (Shah trails by 0.38% on average), the simulations consistently yield transfer-vote margins >8% for Shah, reflecting support from Bellows, Pingree, and King supporters. Adding the two together, Shah wins 97.4% of all Democratic gubernatorial primary simulations while Jackson wins 2.6%. In the chart below, each point represents a simulated election drawn from a preference model estimated using the UNH data.

Given the lack of prior analysis beyond first-choice polls, prediction betting markets are among the few existing indicators of the 2026 Maine primary outcomes. Compared to our simulations, the markets are bullish on a Jackson victory. At the time of writing, the Polymarket opens in a new tab and Kalshi opens in a new tab markets both give Jackson an approximately 60% chance of winning compared to approximately 30% for Nirav Shah.

Below, we provide example simulations that led to Shah and Jackson victories (The exact numbers in the example represent the random draws used in this particular run of the simulation; in practice, we average over many such random runs rather than relying on any one.). The two simulations are highlighted in the scatter plots above.

Example Shah Victory In the example Shah victory shown below, Shah and Jackson begin with nearly equal first-choice shares, but in each elimination round, Shah receives a greater share of the vote transfers, especially following the eliminations of King and Bellows.

Example Jackson Victory In all simulated Jackson victories, Jackson begins with a sufficiently large first-choice lead to offset vote transfers to Nirav Shah in elimination rounds. In the example shown below, Jackson leads Shah by over 13% in first-choice votes. When Hannah Pingree is eliminated in Round 4, 45.5% of her votes are transferred to Shah (compared to 15.5% to Jackson), but this transfer is insufficient to offset Jackson’s lead.

While the simulated elections are grounded in the May UNH poll, sentiment may still shift considerably ahead of the June 9 primary, particularly given the recent cross-endorsements. The cross-endorsement could reduce Shah’s vote-transfer margin from Bellows and Pingree. Further, we also find strong support for Jackson among younger voters. Over half (54.5%) of respondents aged 18-34 ranked Jackson first in the poll, compared with only 25% across all other age groups combined, reflecting a +29.5% gain among youth, as shown in the table below. Maine has one of the highest youth turnout rates in the nation opens in a new tab, ranking second in 2024 behind only Minnesota, so a further increase in youth turnout in 2026, combined with further cross-endorsement impact, should narrow Shah’s lead come June 9.

Democratic Primary for ME-02

In ME-02, four democratic candidates are competing to face off against former Maine governor Paul LePage in the November general election. While Democrat Jared Golden currently holds the seat, Maine’s northern district has increasingly leaned conservative in recent years. In fact, the district is currently the Democratic-held House seat with the highest level of support for Trump in 2024 opens in a new tab.

The candidates are Joe Baldacci, Matthew Dunlap, Paige Loud, and Jordan Wood. Baldacci and Dunlap have long held political office in Maine, with Baldacci currently serving in the Maine State Senate and Dunlap currently serving as State Auditor. In contrast, Loud, a social worker, and Wood, a PAC executive campaigning on campaign finance reform, are seeking their first terms in state office.

The UNH poll shows a close race between Wood and Baldacci for first-place votes, with Wood leading by 1.2%. With nearly a quarter of respondents still undecided ahead of the primary, ME-02 remains highly contested.

The second-choice selections among UNH poll respondents show that Baldacci has greater support than Wood among rivals. Notably, among those ranking Dunlap first, almost twice as many (34.9%) rank Baldacci second compared to Wood (19.3%).

Forecasting the ME-02 Democratic Primary

Simulations of the ME-02 Democratic primary illustrate the importance of Baldacci’s lead in vote transfers. Overall, Baldacci wins in 67% of simulations, Wood in 20.5%, and Dunlap in 12.5%. Unlike the gubernatorial primary, the final pair of candidates varies across simulations and has a large influence on the election outcome. The scatter plots below group simulations based on the final pair of candidates. In a majority of the simulations (73.4%), Baldacci and Wood are the final two candidates. Across these simulations, Wood has a 0.59% lead in first-choice votes, but Baldacci has a 4.99% lead in vote transfers, giving Baldacci the advantage overall. The two other notable final pairs are Baldacci-Dunlap and Wood-Dunlap, both of which are less competitive. In the 15% of simulations that reach a Baldacci-Dunlap final round, Baldacci frequently wins, primarily due to an average 4.33% lead in first-choice selections. And then, in the 10.2% of simulations that reach a Wood-Dunlap final round, Dunlap wins handily, dominating Wood in vote transfers with an average lead of 11.49%. We note that in rare cases (1.4%), Paige Loud is a final candidate in our simulations but is defeated by Baldacci or Wood in all such simulations.

In contrast to the gubernatorial primary, the results of our simulations of the ME-02 Democratic primary are largely similar to the odds in betting markets. On both Polymarket opens in a new tab and Kalshi opens in a new tab, Joe Baldacci has a 66% chance of winning, at the time of writing, and Baldacci wins 67% of our simulations.

Below, we provide examples of Baldacci, Wood, and Dunlap victories.

Example Baldacci Victory The example below is an instance of the most common final pair of Baldacci and Wood. As is common for many simulations with Wood and Baldacci as the final pair, Wood begins with a 5% lead in first-choice share over Baldacci, but the initial lead is insufficient to offset the disparity in transfer to Baldacci following Dunlap’s elimination.

Example Wood Victory The example Wood victory below is consistent with nearly all Wood victories in the scatter plots, with Wood’s first-round margin holding up to vote transfers to Baldacci. In the example, Wood begins with a 9% lead in first-choice share and narrowly defeats Baldacci 41.4% to 41.1% in the final round.

Example Dunlap Victory The Dunlap victory below is an instance of a Dunlap-Wood final round. All simulations featuring this final round, including the one below, feature initial Wood leads in the first round but strong vote transfers to Dunlap in elimination rounds. In the example below, while Wood begins with a 8.7% lead in first-choice share, Wood gains only 6.1% through vote transfers, whereas Dunlap gains 15.8%, over twice as much, primarily from Baldacci’s elimination.

Simulation Methodology

We simulate instant-runoff elections by using the UNH ranked-choice poll data to estimate a preference model of ranked choices, and sample elections from that model. We use the contextual choice model of Awadeldkarim et al. [2023] opens in a new tab¹, which accounts for strong correlations in candidate co-occurrences in ballots, such as cross-endorsements. To sample (simulate) an individual hypothetical election, we first sample a choice model from the posterior over choice-model parameters given the ranked polling data. We then use the choice model instantiation to sample ballots for the hypothetical full election and use the instant-runoff voting algorithm, which Maine opens in a new tab uses. For the ME-02 primary, we sample 50,000 ballots per election (the last primary in 2018 opens in a new tab had 45,211 voters), and for the Democratic gubernatorial primary, we sample 100,000 ballots (the 2018 Democratic gubernatorial primary opens in a new tab, the most recent primary with multiple candidates, had 132,250 voters). The main source of fluctuations in outcomes is not in the sampled ballots from a given model, but in the sampling of the model instance from the model posterior. In both competitive primaries, we sampled 1,000 model instances, leading to 1,000 simulated elections for each primary.

The UNH Democratic gubernatorial poll spanned 506 voters, while the Democratic ME-02 poll spanned 238. Of note, as described by the UNH Survey Center (paragraph four of the “Pine Tree State Poll Methodology” section of the report opens in a new tab), the UNH polls are weighted to be representative of the US census population of Maine and do not include any modeling of voter turnout. As such, all statements about forecasting should be interpreted more as statements of candidate support in the Maine population rather than specific statements about ballot box outcome. For more details or for questions, please contact David Liu at davidliu@cornell.edu opens in a new tab.