This commentary is by Geoffrey Skelley, chief elections analyst for Decision Desk HQ, which provides independent results and analysis for U.S. elections.
Can we trust the polls? As an election analyst, I often get this question. That’s understandable. Polls underestimated Donald Trump’s support in the 2016, 2020 and 2024 presidential elections. This November, Democrats are clearly favored to capture the U.S. House of Representatives and have a realistic shot at flipping the U.S. Senate.
The answer? Yes, we can trust the polls — but only if we allow for uncertainty. That means not relying on one poll, but looking at them in aggregate, which usually means looking at a polling average. After all, polls are snapshots in time, so one poll alone can only gauge support at a particular moment.
Every poll will have some degree of polling error, some larger, some smaller. Surveys are statistical instruments that measure public opinion, so they come with mathematical uncertainty. If a poll were fielded 100 times, a candidate’s support would fall within a certain range — say, plus or minus 3 percentage points — in 95 cases. Yet in a close race, that could mean a candidate is leading or losing by a small amount.
By aggregating polls into averages, we can get a better sense of what the universe of polls says about a given race. This reduces the sway of a single survey, particularly an outlier, on our understanding of a race. That’s not to say averages are perfect: They rely on the quality of their inputs, and as we’ve seen with Trump, particularly in 2016 and 2020, a systematic polling error can consistently underrate one candidate. But averages still give us a better sense of the broad electoral picture, including trends regarding who may be gaining or losing support.
However, averaging polls is less helpful when you only have a few surveys, as in Vermont. How do we make sense of Vermont’s gubernatorial contest when we only have three publicly released polls conducted since mid-September?
With so little polling data, we should be even more cautious about drawing conclusions. But we can also consider other data, such as past election results, and other important factors, including national political conditions and incumbency. This can be done quantitatively with a forecast model that gives a win probability for a candidate, or by a more qualitative approach that relies on categorical ratings such as “tossup” or “lean Republican.”
The polls suggest an edge for the Republican incumbent, Gov. Phil Scott, but one small enough that his Democratic challenger, Amanda Janoo, could win. The most recent poll, an early-October survey by State Navigate, found Scott ahead by just 3 points among likely voters, 46% to 43%. This finding was well inside the poll’s margin of error and fell between two earlier surveys, each conducted in September. Most strikingly, a University of New Hampshire Survey Center poll found Janoo up by 6 points, 49% to 43%. Around the same time, a Braun Research poll for Vermont Public found Scott ahead by about 10 points, 44% to 34%.
A simple average of these polls gives Scott about a 2-point edge, 44% to 42%, while an approach that fits the data based on each survey’s date, weighting recent data more heavily, would have Scott up by closer to 5 points, 43% to 38%. Either way, that’s a fairly close race, but again, based on limited data.
Although Vermont is a reliably Democratic state in federal elections, Scott has an impressive electoral record, having won at least 70% of the vote in back-to-back races. All in all, the 2026 electoral environment is very good for Democrats. But it remains unclear if things are so pro-Democratic that a fairly popular incumbent like Scott could lose.
