In Vermont politics, late-summer polling has a habit of making Gov. Phil Scott look vulnerable — right up until election night proves he isn’t.
Two new September surveys appear to tell different stories about this year’s governor’s race. One, published by the University of New Hampshire, gives Democratic challenger Amanda Janoo a six-point lead, which falls in the survey’s margin of error. Another, by Vermont Public, puts Scott up by nine points
Underneath that 15-point difference is a striking similarity. Both polls have Scott garnering almost exactly the same percentage of the vote.
UNH has him polling at 42%. Vermont Public has him at 43%.
Despite the difference in margins, Harvard University government professor and polling expert Chase Harrison called the polls “very good” and “comparable.”
The difference between the two polls appears among Janoo voters and those who remain undecided. UNH’s poll showed 6% of voters were undecided, while Vermont Public’s showed 20%. The Vermont Public poll seems more typical of a late-September survey, Harrison said, as voters have six weeks to make their decision.
Scott has a history of finishing considerably stronger than his autumn polling suggests. Since Scott first ran for governor in 2016, pre-election polls examined by VTDigger have placed his support at an average of about 14 percentage points below the share of the votes he ultimately received. His opponents, by comparison, have finished much closer to their polling numbers.
That history raises the central question behind this year’s conflicting polls: Is Janoo actually making this the closest governor’s race Vermont has seen in years, or is Scott — the country’s most popular governor — once again entering October with support that polling has yet to capture?
Scott’s campaign spokesperson Emma Shouldice wrote in a statement last week that polling has “always undervalued” support for the Governor.
There is evidence for that argument, although the size of the gap has varied considerably from election to election.
The pattern dates to Scott’s first run for governor. An October 2016 Vermont Public poll showed Scott leading Democrat Sue Minter by just one point, with 14% of voters undecided. Scott won by nearly nine points, an eight-point wider margin than the poll suggested.
The gap was negligible two years later, when Vermont Public had Scott 14 points ahead of Democrat Christine Hallquist and he won by nearly 15. It widened considerably after that. In 2020, Scott’s 32-point lead over David Zuckerman became a 41-point victory. Two years after that, UNH had Scott 17 points ahead of Democrat Brenda Siegel in September. He won by 47, a 30-point difference.
The pattern continued in the last gubernatorial election in 2024. UNH’s late-October poll that year had Scott 39 points ahead of Progressive Esther Charlestin. He won by nearly 52 points, about 13 more than the poll indicated.
Past polling misses do not necessarily tell us what will happen this year. Pollsters say surveys measure voter opinion at a particular moment. They do not predict how undecided voters will break, who will turn out or how the electorate may change before election day.
UNH Survey Center director Andrew Smith and Vermont Public pollster Richard Clark told VTDigger in separate interviews that they were confident in the results of their polls.
When looking at the two surveys side-by-side, Harrison said it appears that when voters are “a little less undecided,” they opt for Janoo over Scott. That is different from what Vermont has seen in recent gubernatorial elections, when Scott has ultimately won by large margins.
“If Scott is seen as a competent incumbent, that can carry an awful lot of weight,” Harrison said. “If you’re seen as a competent incumbent who’s a member of an unpopular political party, you often see that weight — the weight of that incumbency — in people saying they don’t know or not turning out.”
So how can two reputable polls conducted at roughly the same time produce such different pictures of the same race?
One possibility is simply how the polls find and question voters.
Both used probability-based sampling, or randomized methods to select respondents from an entire population, rather than letting people opt in. Clark’s poll for Vermont Public uses a random sample of phone numbers. Respondents who don’t pick up phone calls are offered a text-to-web survey.
The UNH poll uses those techniques to recruit a panel of respondents who take repeated polls. Smith said UNH makes constant changes to the panel to reflect population changes. Respondents to this poll may be “slightly more engaged voters,” Harrison said.
The surveys also asked their gubernatorial questions differently.
Vermont Public’s gubernatorial election question came sixth in the survey, according to its methodology report, after questions about approval of President Donald Trump and Scott and how likely respondents were to vote in November.
UNH asked the question later, following a series of questions about voter likelihood.
Vermont Public’s respondents were “primed strongly to think of job performance,” whereas any such effect in the UNH poll would have been “much weaker,” Harrison said.
There may also be something different about the political environment this year. Negative partisanship is playing a significant role in the election, Smith said. The latest poll from NPR, PBS News and Marist found that 80% of Democrats say their vote for Congress is a vote against Trump.
“It’s not that you like your candidate so much, but you dislike the other party’s candidates more than you dislike your own,” Smith said. “The dynamics of this election, both with the way that midterm cycles work, with Trump’s unpopularity, the nationalization of it is what I think is going on, and why you’re seeing Democrats doing much better.”
That could matter particularly for Scott, a Republican who has repeatedly won broad support in heavily Democratic Vermont.
Pollsters nationally have struggled for years with underestimating Republican support. State-level misses in the 2016 presidential election led pollsters to reevaluate their weighting methods. Polling performed better in the 2018 midterms, but another round of misses in 2020 sent pollsters looking for additional explanations.
One might be that Republicans are less likely to participate in polls, Smith said. Pollsters sought to eliminate that bias in 2024 but were not entirely successful.
“Polling is not a statistical science; it’s a psychological science,” Smith said.
UNH and Vermont Public both weight their surveys using U.S. Census data, attempting to match respondent demographics to state-wide proportions for more accurate representation. UNH also uses previous presidential election data, a methodological difference Clark said could help explain the divergent results.
Whatever explains the difference between the two September polls, neither pollster thinks readers should treat his survey as a forecast of the November result.
Clark said pollsters “are not oracles,” and polling doesn’t claim to be predictive.
“I like to say in a poll it’s more like cutting with a chainsaw than cutting with a scalpel,” Clark said. “It is just not as precise as people would like to believe, but it’s also not as bad as people would like to believe.”
And there are reasons this year could break differently from Scott’s previous races. Clark said the race could be close with the result depending in part on voter turnout and enthusiasm for Janoo compared with voters motivated primarily by anger at Trump.
Age could matter, too. Older people tend to vote for Scott and head to the polls out of habit, Smith said. The question, he said, is whether “energized” young voters will show up as well.
That leaves the biggest question unresolved in both September polls: not where Scott’s support stands, but what happens to everyone else over the final weeks of the campaign.
So, what do polls mean? And why do they matter?
For social and political scientists, polls can provide insight into political preference.
“I want to know if a vote for Phil Scott is connected to a vote for Becca Ballant, and who are the ones who are voting for both?” Clark said. “We’ve asked questions about healthcare and the provision of healthcare, how that factors into one’s political preference.”
Clark said the public focus on the polling “horserace” can obscure what voters are actually deciding.
“We get so caught up in all of that we forget that we’re electing somebody to lead our state,” he said. “We’re electing somebody to represent us in Congress. That’s what really matters. Let’s talk about their opinion.”
