Every October, most conversations in my personal and professional life turn to some version of: “Do you think the polls are right?”My answer is: It depends on what you mean by “right.”I’ve run polling operations for campaigns from city council to president. In 2020 and 2024, I ran the polling and analytics programs on the Biden and Harris presidential campaigns. In each race, I used polls not so much to predict who would win, but rather to tell me how the candidate can win. Polls are often accurate in identifying what voters want, but they’re not always right about the outcome of a close election.

The president’s party almost always loses ground in a midterm. This year, that trend is amplified.

Let’s start with the narrowest version of the question: Are polls accurately capturing support for a candidate in a specific election? The most common argument I hear is that “the polls were wrong last time, so they’ll be wrong again this time.”It’s true that most public polling overestimated Democratic support in the past three presidential elections — though the misses were much smaller in 2024 than in 2016 and 2020.

In the past two midterms however, polling was quite accurate and did not overestimate one party or the other. And in recent off-year elections like 2023 and 2025, polls actually underestimated Democrats. Could a systematic polling miss overstate Democratic support this year? Absolutely. But recent history doesn’t give us any reason to assume that will be the case.

Joe on midterms optimism: ‘very doable’ for Dems to do well October 26, 2026 One reason I trust the general direction of current polling is because it aligns with broader signals. The president’s party almost always loses ground in a midterm. This year, that trend is amplified. Voters both in the U.S. abroad are increasingly fed up with incumbents. Trump’s approval ratings are hitting all-time lows. Economic sentiment among Americans is grim, with gas more than $4 a gallon. And recent voter registration primary turnout trends point to Democratic enthusiasm.

Polls can be wrong, and often they are. But right now, they square up with everything else we know.

Another reason for confidence in the polls is that the industry continues to innovate. This cycle, for example, many pollsters are using improved methods to incorporate vote recall, which helps samples better reflect the actual electorate. These advancements aren’t a silver bullet but they reflect how the polling industry learns from its misses.

Because of the growing difficulty to get voters to complete polls via phone, much of the industry has migrated to cheaper, faster, opt-in online panels. However, these panels can be hard to get right.

At the same time, there remain some very serious problems with polling. Most of them stem from the fact that few people are willing to take polls. Those who do participate tend to be quite different from those who do not, in ways that are hard to measure and correct for. This creates a hurdle pollsters call nonresponse bias.

Because of the growing difficulty to get voters to complete polls via phone, much of the industry has migrated to cheaper, faster, opt-in online panels. However, these panels can be hard to get right. Every poll relies on assumptions about who will turn out to vote, and convincing a truly representative slice of the electorate to opt into an online panel presents its own challenges.

This high-frequency, low-cost online polling environment has also opened the door to bad actors. Already this cycle, we’ve seen artificial intelligence-generated respondents, intentionally skewed results and fabricated polling data. Because the industry is unregulated, these bad polls can spread rapidly on social media and often make their way into official news reports and forecasts. Meanwhile, prediction markets have added financial incentives to publish false data, in hopes of moving those markets. All of this has made it harder than ever to trust any single poll.

Democrats appear to hold narrow edge on the Senate: Sabato’s Crystal Ball The biennial debate over whether we can “trust” the polls is frustrating because it obscures the real value of polling data. Across the political ecosystem, we are using polls to understand what motivates voters, what they want for the future and how their views are changing. We are using polls to allocate resources and target persuadable voters. Polling is great at this. It remains one of our very best tools for testing assumptions and listening to the people who will actually decide an election.

What polling is not so good at is telling us whether a candidate will win 49.8% or 49.6% of the vote. Polls capture a moment in time; they don’t predict the future. All polls have statistical uncertainty that is often larger than the margin of a close race.

Here’s how I’ll be reading the polls between now and November: I’ll look at polling averages and trend lines, instead of zeroing in on any single poll. I’ll remain skeptical of polls from new outfits and those that don’t share their methodology. When polls present contradictory information, I’ll look at their demographic and methodological assumptions to see which are plausible and to understand the impact of turnout assumptions. And I’ll be much more interested in what’s motivating voters — whether that’s Trump, data centers, the cost of groceries or something else — than the headline results.

Election night wouldn’t be exciting if we already knew the ending. This year, I encourage you to take polls for what they are: a strategic compass, a trendline, a window into voters’ feelings and motivations — not a prediction of the outcome.

The post I oversaw polling for Biden and Harris. Here’s how I’m reading the midterm polls.