Why Polling Still Works

Nov. 17, 2026
Photo of pins that say VOTE
Adrienne Bresnahan / Getty

Ashley Koning ’09GSAS is the director of the Eagleton Center for Public Interest Polling and an assistant research professor at Rutgers University, America’s oldest university-based, statewide public polling operation. We spoke with her about the 2026 midterm elections, the state of modern polling, and why polling is integral to our “small-d” democracy. 


Tell me about your career path. How did you become interested in polling? 

I credit two courses with inspiring my love of polling and eventual career choice. As an undergraduate studying political science at Franklin & Marshall, I took a course in mass behavior and realized that the key to understanding the political theories I was learning about was data. Then, during my master’s program at Columbia, I took Bob Shapiro’s quantitative methods course and realized how important math — particularly statistics — is to understanding human behavior in the political landscape. I went to Rutgers for my PhD, planning to focus on women in politics. But largely because of the quantitative analysis I had learned at Columbia, I decided to focus on polling. I worked as a graduate assistant in the polling center, where I am now director. 

Why is polling important? 

The modern father of scientific polling, George Gallup, used to say that polling makes for a truer democracy. And I think that’s right, no matter what side of the aisle you’re on. It connects the public with the policymakers that they elect, especially for issues that may never get their day at the ballot box. It drives the narrative of the election cycle, it impacts what media outlets report on, and it impacts how the public perceives the candidate. It can significantly affect fundraising, as well as voter turnout. 

How is it possible to poll effectively in 2026, an era when no one picks up their phone?

Yeah, let’s be honest: It’s an absolute challenge, especially within the last ten years. It used to be that a majority of people would respond to calls on landline phones and complete the survey. Now we’re dealing with response rates in the low single digits — realistically, one to two percent. In addition to changes in how we communicate, there’s a real lack of trust in polling as an institution among certain segments of the population, particularly Republicans and Trump supporters. We work continually on trying to reach people in effective manners, whether by moving toward mixed-mode designs (combining more than one data collection method into a single survey), calling landlines, texting cell phones, inviting people to take online surveys, or sending out postcards and letters. Polls will never be a perfect picture of public opinion, but our job is to estimate what people think at the moment we reach them, and research has shown that low response rates on their own don’t make a poll inaccurate. Polls are still the most systematic tool we have for understanding where the public stands.

How does weighting work in polling? What does it mean? 

Weighting happens at the end of our data-collection process. Essentially, we’re trying to make the sample of people that we talked to look more like the target population by adding more statistical “weight” to demographics who are less likely to respond to polls. If we’re polling everyone nationally, for example, we know some basic things from the US Census — the number of men versus women or how many people fall into different age groups. Then, we adjust the whole sample to match those benchmarks, which gives more weight to responses from groups that are harder to reach, such as young people.

Nowadays, there’s a lot of talk about weighting by party or by ‘recalled vote’ — adjusting the sample so the share of people who say they voted for each candidate in the last election matches the actual result — because polling has shown that we’ve been getting fewer Republicans and particularly fewer Trump-supporting Republicans in our samples. So we have been trying to balance things to give appropriate, representative weight to those Republicans. This is in some ways a subjective part of the process; we have to make judgment calls about which characteristics matter most for making our sample look like the population, and how much to adjust for them. And those judgment calls can cut both ways. Some of the adjustments pollsters made after 2016, 2020 and 2024 to find more Republicans may have contributed to underestimating Democrats this year, in a midterm where Democrats were especially motivated. In many ways, polling is both an art and a science.

How does polling differ in presidential years from midterm and special-election polling? 

The last ten years have been a very interesting era in politics, because we have seen disparities between when Trump is at the top of the ticket versus when he isn’t. Polls have underestimated Republican support during presidential election years, whereas midterm polls have been more accurate. In this last cycle, though polls and polling averages did meaningfully underestimate Democrats, some of the most credible pollsters were right on the mark in many races. Furthermore, I think there were clear signs that we were heading toward a blue tsunami. Every indicator, whether it was issue polling that showed low presidential approval ratings and low ratings on the economy, or pre-election polling on particular races, consistently pointed to a big win for Democrats. 

What is the difference between issue polling and pre-election polling? Do your methodologies differ? 

That’s such an important question, because there is a very big difference between the two. With issue polling, which asks people their opinions on specific issues, such as the economy, we know who the population is. We’re either talking to adults in a target population in a certain geographic area or to registered voters, and we have reliable benchmarks for both from the census and state voter files. But when it comes to pre-election polling, which asks about particular races, it gets much trickier, because we don’t know who is actually going to vote. You’re talking about a population that doesn’t exist yet. One likely reason polls underestimated Democratic performance this cycle is that Democratic voters came out in unprecedented numbers, so the surge was bigger than we predicted.

Senator-elect Annie Andrews of South Carolina felt like a real surprise win for Democrats. But after the election, there were reports that she was confident about winning from internal polling, which she didn’t release. Can you explain the difference between internal and public polls? In this example, why would she have chosen to keep a winning poll quiet? 

This is a great example! Campaign or internal polling is very different from issue polling and pre-election polling done by academic institutions or media outlets. In campaign polling, the focus is the candidate and the opponent. What are the candidate’s strengths and weaknesses? What are the opponent’s? What is playing well? Say the opponent hates puppies and rainbows. How would you, the public, feel about that? Those are the kinds of things that campaign polling is trying to get at. 

Campaigns leak their internal polls when they think it will be advantageous, as polling can have a huge impact on fundraising and turnout. In the example of Annie Andrews, if she had strong internal polling, she likely did not want to release it because she didn’t want Republicans to infuse last-minute money into her opponent’s campaign. And she didn’t want her own supporters to get lazy: Elections depend on turnout, and it can be helpful for candidates to be seen as an underdog. But by the same token, candidates don’t want to be seen as too far behind, because then donors could jump ship. It’s a very delicate balance. 

Over the last few years, there has been an uptick in gambling on elections. How does that impact your work in polling? 

The betting markets have become a very hot topic in polling circles, with some even asking whether these kinds of prediction markets will replace polling. But ultimately, the markets are responding in large part to the information that polling gives them, and the two are answering very different questions — markets answer the who, while polls answer the why. So you can’t have betting markets without polls. Additionally, betting markets are — to use a trendy political term from the last few years — very vibes-based; they reflect what the people placing bets believe, not what a representative sample of the public thinks. Polling is a real scientific and methodological tool that is based on almost a century of research and practice. 

Read more from Rebecca Shapiro